Video processors, methods, devices, storage media, and program products

Abstract

This application provides a video processor, method, device, storage medium, and program product. In this solution, after the presentation layer receives the special effects style configuration, the algorithm engine layer immediately outputs a high-fidelity special effects description file based on the target special effects style and the matching physical and mathematical model of the basic video, improving the accuracy of special effects parameters. The rendering adaptation layer dynamically selects the optimal rendering engine according to the resources of each rendering engine, shortening the rendering time per instance. The two layers are decoupled, allowing description file generation and rendering execution to proceed in parallel. Batch description files can be shared by multiple preview positions once they are ready, and rendering resources are only invested when there is a real preview requirement, avoiding idle time. The presentation layer, algorithm engine layer, rendering adaptation layer, and rendering layer work in sequence to form a compact "accurate generation - fast rendering - instant playback" chain, significantly reducing the user's waiting and repeated export steps in the subsequent video editing stage, and improving the overall video editing efficiency.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a video processor, method, device, storage medium and program product. Background Technology

[0002] In the daily operations of short video, live streaming, and online editing platforms, dynamic effects are often added to enhance video visuals and increase their appeal. Previewing videos with added dynamic effects has become a core element in improving user creation efficiency. Traditional video production requires users to manually edit dynamic effects on a target video using a video editor. During editing, users can view the effect frames being edited for the current frame. After editing the entire target video, it is exported and played back, allowing users to view the overall dynamic effects. If users are not satisfied with the dynamic effects, they need to re-edit the target video, which significantly impacts editing efficiency. Summary of the Invention

[0003] Unlike traditional linear effects production workflows that require complete editing of the video before exporting and playback, this application provides a video processor, method, device, storage medium, and program product with a system architecture that decouples the generation and rendering of description files for effect reference videos. Traditional methods require manual adjustment of effects frame-by-frame within a video editor, and the overall effect can only be viewed after exporting the entire video at once. If the effect is unsatisfactory, it needs to be re-edited, repeatedly rendered and exported, creating a cycle of editing, waiting, and reviewing, resulting in low video editing efficiency and long processing times. This application integrates the batch generation of description files for various effect reference videos and an instant preview function into a single interface. It batch generates description files for multiple effect preview videos under different effect styles and renders them in real time when the user needs a preview. This allows the user to determine whether to select the target effect from the target effect reference video for subsequent video editing based on the preview effect, significantly shortening the effect selection time, reducing repeated export operations, improving video editing efficiency, and solving the core problems of traditional workflows, such as the inability to preview effect video effects and the need for re-editing if unsatisfied.

[0004] This application provides a video processor, including: a presentation layer, an algorithm engine layer, a rendering adaptation layer, and a rendering layer, wherein the algorithm engine layer and the rendering layer are decoupled; the presentation layer is used to display a special effects configuration interface, which includes special effects style configuration items and a special effects reference video display area, the display area including multiple preview slots for displaying special effects reference videos; the algorithm engine layer is used to respond to configuration operations through the special effects style configuration items, obtain a target special effects style; from multiple mathematical algorithm models, call a target mathematical algorithm model adapted to the target special effects style and the base video to generate multiple description files of target special effects; and fuse the description files of multiple target special effects and the description file of the base video to obtain multiple special effects reference videos. The system consists of a description file, with each effect reference video having a corresponding preview position; a rendering layer, which responds to preview operations for the target preview position by sending a rendering service request to the rendering adaptation layer, where the target preview position corresponds to the description file of the target effect reference video; a rendering adaptation layer, which responds to the rendering service request by selecting a target rendering engine from multiple rendering engines and returning the target rendering engine's identification information to the rendering layer; and a rendering layer, which, based on the target rendering engine's identification information, invokes the target rendering engine and renders the target effect reference video to the target preview position according to the description file of the target effect reference video, allowing the user to determine whether to select the target effect from the target effect reference video for subsequent video editing based on the preview effect.

[0005] This application also provides an electronic device, including: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps in the above method.

[0006] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, causes the processor to perform the steps in the above-described method.

[0007] This application also provides a computer program product, including: a computer program that, when executed by a processor of an electronic device, causes the processor to perform the steps in the above-described method.

[0008] The technical solution provided in this application embodiment, after the presentation layer receives the special effects style configuration, the algorithm engine layer immediately calls the matching physical and mathematical model according to the style description information to output a high-fidelity special effects description file, improving the accuracy of special effects parameters; the rendering adaptation layer dynamically selects the optimal rendering engine according to the resources of each rendering engine, shortening the rendering time of a single instance; the two layers are decoupled, so that the generation of description files and the rendering execution are parallel and do not conflict, and batch description files can be shared by multiple preview positions once they are ready, and rendering resources are only invested when there is a real preview need, avoiding idle running; the presentation layer, algorithm engine layer, rendering adaptation layer and rendering layer take turns to form a compact link of "accurate generation-fast rendering-instant playback", which significantly reduces the waiting and repeated export steps for users in the subsequent video editing stage, and improves the overall video editing efficiency. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0010] Figure 1 A schematic diagram of the structure of a video processor provided in an exemplary embodiment of this application;

[0011] Figure 2a A schematic diagram of a video editing page provided for an exemplary embodiment of this application;

[0012] Figure 2b A schematic diagram of the details page of the preview position provided for an exemplary embodiment of this application;

[0013] Figure 3 A schematic flowchart illustrating a video processing method provided in an exemplary embodiment of this application;

[0014] Figure 4 A schematic diagram of the structure of an electronic device provided for another exemplary embodiment of this application. Detailed Implementation

[0015] Continuing with the background technology, traditional video production suffers from at least the following shortcomings:

[0016] 1. Dispersed operation: The traditional workflow breaks down special effects generation, parameter adjustment, rendering, previewing and downloading into multiple independent tools. Users need to repeatedly import and export between software. The learning cost and the probability of error increase linearly with each step, and the overall efficiency decreases exponentially.

[0017] Second, lack of visualization: the special effects parameters are presented in static forms or text lists, and users cannot perceive the adjustment results in real time. They can only "blindly adjust" and wait for the complete rendering, resulting in a long feedback cycle and frequent interruptions to the creative process.

[0018] 3. Weak batch processing capability: Single-instance serial rendering means that only one effect can be processed at a time, and repetitive operations consume a lot of manual time; when multiple versions of the same style need to be compared, the linear working method makes the total time increase exponentially with the number of versions, which is difficult to meet the needs of rapid content iteration.

[0019] 4. Poor cross-platform compatibility: The toolchain is deeply bound to specific operating systems and underlying libraries, which means that team collaboration must use the same hardware and software versions; remote or outsourced processes cannot reproduce the scenario due to environmental differences, and information transfer relies on manually exporting files, resulting in "data silos" and version management risks.

[0020] 5. Dispersed file management: Special effects source files, intermediate rendering results, and final products are scattered in different local directories, lacking a centralized index; as the amount of materials increases, retrieval relies on manual memory, which easily leads to phenomena such as "not found, wrong version used, and repeated rendering", resulting in a double waste of storage space and time.

[0021] 6. Inability to remotely generate and preview in batches: The lack of a network interface in the local tool limits "generate-preview" to a single workstation. Remote rendering cannot be directly driven, and multiple people are forced to repeatedly transfer large files, making batch concurrency and real-time sharing impossible.

[0022] In summary, each iteration in traditional video production incurs full rendering costs, resulting in a triple waste of time, computing power, and manpower, and becoming a major bottleneck for improving video editing efficiency.

[0023] To address the aforementioned technical issues, in this embodiment, after the presentation layer receives the special effects style configuration, the algorithm engine layer immediately calls the matching physical and mathematical model based on the target special effects style and the basic video to output a high-fidelity special effects description file, improving the accuracy of special effects parameters. The rendering adaptation layer dynamically selects the optimal rendering engine based on the resources of each rendering engine, shortening the rendering time per instance. The two layers are decoupled, allowing the generation of description files and the execution of rendering to proceed in parallel. Batch description files can be shared by multiple preview positions once they are ready, and rendering resources are only invested when there is a real preview requirement, avoiding idle time. The presentation layer, algorithm engine layer, rendering adaptation layer, and rendering layer work in sequence to form a compact link of "accurate generation - fast rendering - instant playback", significantly reducing the waiting and repeated export steps for users in the subsequent video editing stage, and improving the overall efficiency of video editing.

[0024] Furthermore, the video processor includes multiple layers that can realize functions such as special effects generation, parameter adjustment, rendering, preview, and downloading. This solves the technical problem of traditional workflows that separate special effects generation, parameter adjustment, rendering, previewing, and downloading into multiple independent tools, requiring users to repeatedly import and export between software. The learning cost and the probability of errors increase linearly with each step, and the overall efficiency decreases exponentially. With this processor, users can preview the target special effects reference video without repeated importing and exporting. The entire process is automated, requires no learning cost, and the probability of errors is greatly reduced, resulting in an exponential increase in overall efficiency.

[0025] Furthermore, the video processor includes a presentation layer. Users can configure special effects styles and preview target special effects reference videos through the presentation layer. The presentation area of ​​the presentation layer includes multiple preview slots for displaying special effects reference videos. Each preview slot corresponds to a details page. The details page contains an area for modifying the description information of the description file and a display area for the target special effects reference video. After modifying the description information in the modification area of ​​the details page, the modified target special effects reference video can be rendered in real time in the display area of ​​the target special effects reference video. This solves the technical problem that special effects parameters are presented as static forms or text lists, which prevent users from immediately perceiving the adjustment results. Users can only "blindly adjust" and wait for the complete rendering, resulting in long feedback cycles and frequent interruptions to creative thinking. Users can immediately perceive the adjustment results of the special effects reference video. Based on this, users can edit subsequent videos better and more skillfully, and the absence of feedback cycles improves overall efficiency.

[0026] Furthermore, the video processor's algorithm engine layer can call the appropriate mathematical algorithm model with a single input, enabling the batch generation of multiple differentiated target effect description files. This solves the technical problems of single-instance serial rendering, which dictates that only one effect can be processed at a time; the rigidity of traditional fixed template parameters; and the significant manual time consumed by repetitive operations. When multiple versions of the same style need to be compared, the linear workflow causes the total time to increase exponentially with the number of versions, making it difficult to meet the needs of rapid content iteration. This eliminates repetitive labor and saves manpower, greatly improving overall efficiency.

[0027] Furthermore, the video processor can adopt a unified web architecture and standardized API interface, encapsulating the entire chain of effects generation, parameter adjustment, rendering, preview, and download within the same browser environment. It does not need to be bound to a specific operating system or local underlying library, solving the technical problems of poor cross-platform compatibility caused by the deep dependence of the toolchain on a single platform and the need for unified hardware and software versions for team collaboration. Users can directly drive rendering in any browser on Windows, macOS, Linux, and mobile devices. Remote and outsourcing processes are no longer limited by environmental differences. Information can be transmitted through description files to reproduce the scene, eliminating "data silos" and version management risks. The overall collaboration efficiency increases exponentially.

[0028] Furthermore, the video processor has a built-in file management layer that automatically and centrally stores description files and preview videos according to the effect style and date level, and updates the searchable index in real time. This solves the technical problems of effect source files, intermediate results, and finished products being scattered in different local directories and lacking a centralized index, which leads to "not being able to find, using the wrong version, and repeated rendering". Users can locate the required description file in seconds through the display layer. The system automatically cleans up expired files, saving both storage space and retrieval time, and significantly reducing management overhead.

[0029] Furthermore, the video processor exposes standardized rendering APIs through a unified web portal, so that "generation-preview" is no longer limited to a single workstation. Remote users can initiate rendering requests in batches from any terminal, and the system instantly returns a lightweight description file and renders the first frame in real time. This solves the technical problem that local tools lack network interfaces and cannot remotely generate and preview in batches. Multi-person collaboration can achieve concurrent preview and real-time sharing without repeatedly transferring large files. The batch concurrency capability increases exponentially, significantly shortening the content iteration cycle.

[0030] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] Before providing a detailed description of the technical solutions provided in the various embodiments of this application, the following is a brief explanation of several technical concepts involved in this application:

[0032] RESTful API: A web application interface that conforms to the REST architectural style and supports standard HTTP methods for resource manipulation.

[0033] Hover preview (or hover operation): A front-end interaction design technology that allows a preview video to play automatically when the user hovers the mouse over an effect card (or preview area).

[0034] Batch operation (or batch generation): Supports the function of simultaneously performing operations such as generating, previewing, and downloading multiple special effects files.

[0035] Dynamic index cache (or index): A file indexing system maintained in real time, providing fast retrieval and metadata query capabilities. Through the interface, it can read special effects and videos generated on the server in real time.

[0036] Responsive design: a front-end design technique for web interfaces that adapt to different screen sizes.

[0037] Kdenlive: An open-source, cross-platform non-linear video editor or video processor that supports multi-track editing, filters, transitions, keyframe animation, and other functions. It uses the MLT framework as its rendering engine, and effects are described in XML files.

[0038] MLT Multimedia Processing Framework: Media Lovin' Toolkit is an open-source multimedia processing framework that provides features such as multi-track editing and keyframe rendering.

[0039] XML effects templates: Extensible markup language files used to describe video effects parameters, keyframe animations, filter effects, etc., which can be directly loaded and used by Kdenl ive.

[0040] Physical / mathematical models: refer to the algorithmic models used to generate special effect parameters, such as damped vibration formulas, Bezier interpolation, Gaussian blur, etc.

[0041] Figure 1 This is a schematic diagram of the structure of a video processor provided for an exemplary embodiment of this application. Figure 1 As shown, it includes: Presentation Layer 1, Algorithm Engine Layer 2, Render Adaptation Layer 3, and Rendering Layer 4, with Algorithm Engine Layer 2 and Rendering Layer 4 decoupled. The video processor can be implemented as a web page, such as the web page of the MLT multimedia processing framework.

[0042] Among them, the display layer 1 is used to display the special effects configuration interface. The special effects configuration interface includes configuration items for special effects styles and a display area for special effects reference videos. The display area includes multiple preview slots for displaying special effects reference videos.

[0043] Algorithm engine layer 2 is used to respond to configuration operations through the special effects style configuration items, obtain the target special effects style; from multiple mathematical algorithm models, it calls the target mathematical algorithm model that is adapted to the target special effects style and the base video to generate multiple description files of target special effects; the description files of multiple target special effects and the description files of the base video are merged to obtain description files of multiple special effects reference videos, and each description file of the special effects reference video corresponds to a preview bit;

[0044] Rendering layer 4 is used to respond to preview operations for the target preview position and send a rendering service request to the rendering adaptation layer. The target preview position corresponds to the description file of the target effect reference video.

[0045] Rendering adaptation layer 3 is used to respond to rendering service requests, select the target rendering engine from multiple rendering engines, and return the identification information of the target rendering engine to the rendering layer;

[0046] Rendering layer 4 is also used to call the target rendering engine based on the identification information of the target rendering engine, and render the target effect reference video to the target preview position according to the description file of the target effect reference video, so that the user can determine whether to select the target effect in the target effect reference video for subsequent video editing based on the preview effect.

[0047] In this embodiment, the presentation layer 1 displays the special effects configuration interface, which includes special effects style adaptation items and a special effects reference video display area. The special effects style adaptation items allow users to configure special effects styles in real time according to actual needs. This enables the generation of corresponding special effects reference video description files based on the configured target special effects style, thereby rendering the corresponding special effects reference video based on the description files. These description files are newly generated, not pre-generated. Special effects styles include, but are not limited to, blur effects, color effects, glitch effects, jitter effects, transition effects, and scaling effects. Furthermore, the number of corresponding special effects reference videos generated based on the configured target special effects style can be one or more, meaning that description files for corresponding special effects reference videos can be generated in batches based on the configured target special effects style. The number of description files generated for different configured special effects styles may be the same or different. The number of description files generated in batches for each special effects style depends on specific needs, and this embodiment does not limit this. The description file can be an XML format description file, but is not limited to this.

[0048] The special effects video showcase area includes preview slots for displaying reference videos of special effects. Typically, each reference video within the same special effects style has its own corresponding preview slot. That is, the number of preview slots in the special effects video showcase area is related to the configured target special effects style, and the number of preview slots is the same as the number of description files or corresponding reference videos to be generated under the target special effects style. In other words, one preview slot is associated with one description file or corresponding reference video of a special effects video, and when a reference video of a special effects video is played, it is played in its associated preview slot. For example, if a blur effect requires 5 corresponding effect reference videos, then when the configured target effect is blur, 5 preview slots will be displayed in the effect reference video display area, with each effect reference video corresponding to its own preview slot; if a shake effect requires 6 corresponding effect reference videos, then when the configured target effect style is shake, 6 preview slots will be displayed in the effect reference video display area, with each effect reference video corresponding to its own preview slot; if a zoom effect requires 6 corresponding effect reference videos, then when the configured target effect style is zoom, 6 preview slots will be displayed in the effect reference video display area, with each effect reference video corresponding to its own preview slot.

[0049] Furthermore, the special effects configuration interface also includes a configuration item for the number of special effects reference videos generated based on the special effects style. This is used to configure the number of special effects reference videos generated under the target special effects style after the target special effects style is configured. Based on the target special effects style and its corresponding number of special effects reference videos, a corresponding number of description files containing special effects reference videos of the target special effects style are generated in batches. Comparing the generation of N special effects versions under the target special effects style, the traditional solution takes approximately N * (single generation + rendering time), while this solution takes approximately (single batch generation time) + (single rendering time * concurrency). When N is large, the efficiency advantage is extremely significant, achieving an order-of-magnitude improvement in batch processing efficiency. Configuring the number of special effects reference videos corresponding to a special effects style through the special effects reference video number configuration item is one implementation method. Other implementation methods for determining the number of special effects reference videos corresponding to a special effects style can be found in the relevant descriptions of the following algorithm engine layer 2 embodiments, which will not be elaborated here.

[0050] It's important to note that when switching configurations for different special effects styles, the preview positions displayed in the special effects reference video showcase will also change accordingly. For example, if the target special effects style is configured as blur and the number of reference videos is configured as 5, then 5 preview positions under blur will be rendered and displayed in the special effects reference video showcase. Furthermore, if the target special effects style is further configured as jitter and the number of reference videos is configured as 6, then 6 preview positions under jitter will be rendered and displayed in the special effects reference video showcase. The 5 preview positions under blur and the 6 preview positions under jitter can be completely different. In other words, although multiple preview positions under different special effects styles can be displayed in the special effects reference video showcase, these multiple preview positions under different special effects styles are different.

[0051] For example, an exemplary diagram of the special effects configuration interface is shown below. Figure 2aAs shown in the diagram, the first area is a configuration area, and the second area is a special effects reference video display area. Further, the first area includes a first sub-area and a second sub-area. The first sub-area contains special effects style configuration items for configuring the special effects style of pre-generated description files for special effects reference videos, so that after configuring the target special effects style, multiple preview positions corresponding to the pre-generated description files for special effects reference videos can be directly displayed in the display area. The second sub-area contains the special effects style configuration items and quantity configuration items mentioned in the above embodiment for generating new description files for special effects reference videos, so that after configuring the target special effects style and its corresponding quantity, the corresponding number of description files for target special effects reference videos can be generated in real time, and the corresponding preview positions can be generated and rendered in real time in the display area. If the preview effect of the special effects reference video rendered based on the description files for special effects reference videos under each special effects style in the first sub-area cannot meet the user's needs, a new description file for special effects reference videos can be reconfigured and generated in the second sub-area, allowing the user to preview the special effects reference videos corresponding to the new description files and select the desired special effects.

[0052] The aforementioned implementations related to the presentation layer, leveraging a lightweight interactive approach of "real-time configuration - instant mapping - preview visibility," can instantly transform the selection of style and quantity into the generation of entirely new description files and the dynamic arrangement of preview positions. This achieves a differentiated effect for horizontal comparison within the same screen, helping users quickly determine the effects they need. The combination of pre-generated and real-time description files, along with a decoupled rendering chain, avoids redundant pre-generated inventory and the time spent waiting for complete rendering, significantly shortening the selection cycle. It fully utilizes the technical advantages of batch generation and remote preview, while simultaneously improving the concurrency and response speed of the video processor, ultimately achieving a simultaneous improvement in video editing efficiency and user experience.

[0053] In this embodiment, the description file of the target effect reference video is obtained by fusing the description file of the target effect and the description file of the base video. Therefore, before generating the description file of the target effect reference video, the description files of the target effect and the base video must be obtained first. The description file of the target effect can be generated using a target mathematical algorithm model adapted to the target effect. Furthermore, the description file of the target effect can be generated using a target mathematical algorithm model adapted to both the target effect and the base video.

[0054] Based on this, the aforementioned algorithm engine layer 2 is used to respond to configuration operations via the special effects style configuration items, obtain the target special effects style; from multiple mathematical algorithm models, it calls the target mathematical algorithm model adapted to the target special effects style and the base video to generate multiple description files for the target special effects; it merges the description files for the multiple target special effects with the description files for the base video to obtain description files for multiple special effects reference videos, where the base video can be a video to be edited by the user later, or a general video. The algorithm engine layer locks the corresponding mathematical algorithm model immediately upon receiving the special effects style configuration operation, batches out differentiated description files and merges them with the base video description files to generate multiple special effects reference video description files at once; the base video can be either the material to be edited or a general sample, so that the same batch of description files can be directly reused in different scenes without repeated calculations, significantly reducing the number of generation times and storage redundancy, simultaneously improving batch generation efficiency and the scheduling flexibility of the subsequent rendering adaptation layer, thereby improving overall efficiency.

[0055] This embodiment does not limit the implementation method of calling a target mathematical algorithm model that is compatible with the target effect style and the base video from multiple mathematical algorithm models to generate multiple description files for the target effect. For example, a knowledge graph-based reasoning and matching method can be used, which involves pre-constructing a "style-algorithm-parameter" triple knowledge graph, with nodes weighted by attributes such as applicable frame rate and amplitude range; using the metadata of the target effect style and the base video as query entities, performing shortest path reasoning along the edge weights of the graph, and automatically locating the most suitable mathematical algorithm node; performing batch random perturbation on the parameter templates associated with the selected algorithm node to generate N sets of differentiated description files. By using a knowledge graph-based reasoning and matching method, the shortest path query can be completed in milliseconds through the "style-algorithm-parameter" triple network and weighted edges, automatically locking in the mathematical model that best matches the target effect and video metadata, reducing the mismatch rate; the graph nodes are accompanied by semantic attributes, which facilitates subsequent backtracking and recommendation, providing a reliable model selection credential for decoupling the algorithm layer and the rendering layer. For example, the parameter evolution method based on genetic algorithms encodes each candidate control parameter set as a chromosome, using the target effect style description and basic video metadata as input for fitness evaluation. The candidate control parameter set refers to a set of specific values ​​carried by each chromosome in the initial generation of the genetic algorithm population, which can be used to describe the target effect. Examples include the coordinates of control points on a Bézier curve, the amplitude, frequency, and attenuation coefficient of damped vibrations. These values ​​cover different values ​​within the allowed range of parameters, constituting candidate solutions in the algorithm's evolution process. After fitness evaluation, these values ​​are continuously selected, crossovered, and mutated until convergence yields the optimal or near-optimal combination of control parameters, used to generate the corresponding effect style description file. Random chromosomes are introduced into the initial population, undergoing selection, crossover, and mutation, followed by iterative evolution until fitness converges, yielding the optimal control parameter set. A Gaussian perturbation is applied to this optimal chromosome to generate N near-optimal solutions, which are directly substituted into the general curve equation for frame-by-frame sampling, outputting N differentiated description files. The parameter evolution method based on genetic algorithms, guided by a fitness function, can automatically converge to the optimal control point set within the search space without a pre-set template library, breaking through the style upper limit of fixed templates. At the end of the evolution, Gaussian perturbation is introduced to generate N sets of near-optimal chromosomes, ensuring that description files under the same style are different in amplitude, period, and trajectory, providing rich contrast materials for the presentation layer. For a more preferred implementation method of generating multiple target effect description files by calling target mathematical algorithm models adapted to the target effect style and the base video from multiple mathematical algorithm models, please refer to the relevant descriptions in the following embodiments, which will not be repeated here.

[0056] This embodiment does not limit the implementation method of fusing the description files of the multiple target effects with the description file of the base video to obtain the description files of multiple effect reference videos. For example, a fusion method based on timeline anchor point injection can be used. Specifically, the base video description file is pre-parsed into an anchor timeline with empty slots, and each anchor point corresponds to the starting frame number where the effect can be inserted. After splitting each target effect description file into sub-segments according to the anchor frame number, the sub-segments are directly written into the corresponding anchor slots to form a multi-track parallel structure. Then, the track header information is merged to generate a reference video description file with multiple effects superimposed. In this way, fusion can be completed without copying the base video frame data. By employing a timeline anchor-injection-based fusion method, the basic video frame data is copied zero-time. Multiple effects can be mounted in parallel simply by filling empty slots, maintaining a constant file size. The fusion process uses precise frame number alignment to avoid time drift. Furthermore, this method decouples the algorithm and rendering layers, allowing previewing immediately after the description file is generated. The rendering layer only needs to read the newly added tracks to output the visuals, significantly shortening the waiting time from batch generation to visual display and fully amplifying the advantages of batch preview. Another example is the differential patch-based fusion method, which treats the basic video description file as a baseline version. Each target effect description file is converted into a differential patch sequence, with each patch recording only the keyframe offset and parameter increment relative to the baseline. The patch overlay engine applies the differentials sequentially, dynamically reconstructing a new set of keyframes containing the effects, and then writing it back to the same base timeline, outputting a reference video description file for the effects after fusion of multiple differential patches, thereby reducing the writing of redundant data. The "same base timeline" refers to using a single, unique base video description file as the time reference throughout all fusion operations. Its track structure, frame rate, total duration, and source material references remain unchanged. Various effect difference data (keyframes, parameters, patches) are only appended or superimposed on this base timeline without generating new base tracks. This ensures that multiple versions of effect reference video description files share the same time coordinate system, avoiding repeated copying of base video frame data and achieving lightweight fusion. The differential patch-based fusion method replaces complete frame writing with keyframe offsets and parameter increments, resulting in a linear, slight increase in storage overhead with the number of effects. Patch sequences are reusable and rollbackable, supporting rapid switching or overlaying of different effect versions. Combined with an on-demand rendering strategy, only a small amount of data needs to be transmitted during remote calls to reconstruct the effect timeline, reducing network bandwidth usage. Simultaneously, differential data provides accurate interpolation for subsequent downgrade rendering and the selection of placeholder videos, ensuring seamless transitions between rendering the first frame and then completing other frames in high definition, further enhancing remote collaboration and real-time preview experiences. For a preferred embodiment of fusing the description files of the multiple target special effects with the description file of the base video to obtain the description files of multiple special effects reference videos, please refer to the relevant description of the following embodiments, which will not be elaborated in detail here.

[0057] Furthermore, after obtaining the description files of multiple special effects reference videos, each description file can be stored in a first designated storage space, such as a local designated storage space or a server-side designated storage space. Simultaneously, first index information is assigned to each description file of the special effects reference videos, associating each description file with its respective first index information for storage. This facilitates the rapid retrieval of the corresponding description file from the first designated storage space when rendering the corresponding special effects reference video based on its description file, using the first index information. The first index information can be, for example, multi-level storage path information, special effects information, or special effects style identification information for the description file of the special effects reference video. This embodiment does not limit this, as long as it uniquely identifies the description file of the special effects reference video. Correspondingly, the preview position corresponding to each description file of the special effects reference video can also be associated with the corresponding first index information. This allows for the immediate and rapid retrieval of the corresponding description file from the designated storage area based on the first index information associated with the target preview position when performing a preview operation based on the target special effects style. Therefore, index information is added to the description file, which can quickly locate the file's storage location, improving file search efficiency. This also facilitates the rapid rendering of special effects reference videos for users to preview and select special effects templates, thereby improving the efficiency of subsequent video special effects editing.

[0058] In this embodiment, when a user has a preview requirement, a preview operation can be initiated on the target preview position under the target special effects style in the display layer 1. Correspondingly, the rendering layer 3 is used to respond to the preview operation for the target preview position by sending a rendering service request to the rendering adaptation layer. The target preview position corresponds to the description file of the target special effects reference video. The rendering layer 3 and the algorithm engine layer 2 respond to different operations, thereby achieving decoupling between the rendering layer and the algorithm engine layer. The preview operation only triggers the rendering layer to request the description file from the rendering adaptation layer and initiate a rendering service request, which is completely separated from the stage of generating the description file in the algorithm engine layer. The two respond independently to different events, so that the rendering resources are rendered when the user has a preview requirement, avoiding invalid rendering occupation, shortening the first frame waiting time, and simultaneously reducing the load on the video processor. With this decoupling, the batch-generated description files can provide sufficient preview materials for the display layer in advance. Users can quickly see the effect of the special effects reference video by clicking on any preview position. The whole process realizes a lightweight loop of "generate first, then render on demand", which significantly accelerates the special effects confirmation process in video editing. Furthermore, because the generation of the description file is decoupled and executed in parallel, the waiting time for users from configuring effects styles to seeing the first batch of previews ready is reduced from minutes, which increases linearly in the traditional approach, to almost constant seconds (or even milliseconds, depending on the description file generation speed), achieving a leap in efficiency. In addition, the decoupled architecture allows the algorithm engine and rendering engine to evolve and expand independently; if a new effects style or rendering engine is added, the entire system does not need to be rebuilt, greatly reducing maintenance costs and upgrade complexity. Moreover, the lightweight description file format enables remote collaboration and cloud rendering, laying the foundation for building a distributed video processor and achieving a significant enhancement in system scalability and compatibility.

[0059] In this embodiment, the rendering adaptation layer 4 is used to respond to the rendering service request sent by the rendering layer 3, select a target rendering engine from multiple rendering engines, and return the identification information of the target rendering engine to the rendering layer. This embodiment does not limit the implementation method of selecting the target rendering engine from multiple rendering engines. For example, the rendering service request includes the current terminal's CPU utilization, remaining battery power, and expected rendering duration; the rendering adaptation layer compares the real-time performance benchmarks (such as FPS / Watt) of each engine with the requested values, automatically assigns the task to the available engine with the highest frame rate per unit power consumption, and forces the task to be downgraded to a software editing engine in low-power scenarios. This method can significantly extend battery life and reduce device heat generation on mobile or edge devices, ensuring a smooth and lag-free remote preview process; at the same time, it maintains a fast first-frame experience, aligning with the overall on-demand rendering and instant visibility rhythm, and improving the friendliness of cross-platform collaboration. For example, if a rendering service request includes user identification information, the rendering adaptation layer queries the engine records and ratings of the user's most recent N successful renderings, prioritizing the engine with the highest historical success rate and marked as "satisfactory." If no record is found, it falls back to the default priority queue. This method can reduce repeated rendering or manual switching caused by individual engine differences, shortening the effects confirmation path. The combination of personalized scheduling and the high reusability of batch description files makes the overall editing workflow more coherent, further improving video creation efficiency. For the preferred implementation method of selecting a target rendering request from multiple rendering engines, please refer to the relevant descriptions in the following embodiments, which will not be repeated here.

[0060] Correspondingly, the rendering layer 3 is also used to invoke the target rendering engine according to the identification information of the target rendering engine, and render the target effect reference video to the target preview position according to the description file of the target effect reference video, so that the user can determine whether to select the target effect in the target effect reference video for subsequent video editing based on the preview effect. Rendering resources are only invoked when the user actually triggers the preview operation, avoiding the need to prepay the full rendering cost for each effect that may be rejected by the user in traditional solutions, and eliminating the problem of idle computation.

[0061] This embodiment does not limit the implementation method of rendering the target effect reference video to the target preview position based on the description file of the target effect reference video. For example, a lightweight first-frame rendering method can be adopted, specifically parsing the description file to obtain the frame rate, resolution, and keyframe sequence, extracting only the parameters of frame 0; calling FFmpeg instructions to generate a single-frame MP4, and the front end immediately plays it using the Video tag (such as setting it to mute, autoplay, or loop playback); the back end puts the complete rendering task into an asynchronous queue, and seamlessly replaces the video source (Src) after the task is completed, thereby achieving fast frame padding and background completion. Another example is the GPU shader real-time rendering method, which converts the keyframe sequence in the description file into a Uniform Buffer; the WebGL / Metal / Vulkan vertex shader interpolates according to the frame number; the fragment shader directly synthesizes the effects and basic video textures; the frame buffer is output to the preview position canvas in real time, without saving the file to disk, the first frame has almost no delay, and it is completed synchronously with the user's hover or click action.

[0062] Furthermore, while rendering the target effect reference video to the target preview position based on the target rendering engine's identification information and the description file of the target effect reference video, the target effect reference video can also be stored in real time to a second designated storage space. This allows the target effect reference video to be retrieved directly from the second designated storage area when previewing it later, eliminating the need for repeated rendering of the same target effect reference video and saving rendering resources. Simultaneously, while storing the target effect reference video in the second designated storage space, second index information can be assigned to it, establishing a relationship between the second index information and the corresponding target effect reference video, as well as a relationship between the second index information and the target effect reference video. This allows users to directly find the rendered target effect reference video based on the second index information when they need to preview it, and then display the target effect reference video in the corresponding preview position for user preview. The first and second designated storage spaces can be the same or different storage spaces. Similarly, the first and second index information of the target special effects reference video and its corresponding description file can be the same or different index information. Furthermore, a relationship can be established between the first and second index information to obtain a centralized index between the description file and the special effects reference video. This facilitates subsequent searching of the description file or special effects reference video based on the centralized index. This solves the technical problems of traditional solutions where special effects source files, intermediate rendering results, and final products are scattered across different local directories, lacking a centralized index. It also addresses the issues of relying on manual memory for retrieval as the amount of material increases, leading to errors such as "not found," "using the wrong version," and "repeated rendering," as well as wasting both storage space and time. The centralized index information allows for quick location of files, improving file search efficiency and facilitating rapid rendering of special effects reference videos for user preview and selection of effect templates, thereby improving the efficiency of subsequent video effects editing.

[0063] The technical solutions provided in the above embodiments of this application, after the presentation layer receives the special effects style configuration, the algorithm engine layer immediately calls the matching physical and mathematical model according to the style description information to output a high-fidelity special effects description file, improving the accuracy of special effects parameters; the rendering adaptation layer dynamically selects the optimal rendering engine according to the resources of each rendering engine, shortening the rendering time of a single instance; the two layers are decoupled, so that the generation of description files and the rendering execution are parallel and do not conflict, and batch description files can be shared by multiple preview positions once they are ready, and rendering resources are only invested when there is a real preview need, avoiding idle running; the presentation layer, algorithm engine layer, rendering adaptation layer and rendering layer take turns to form a compact link of "accurate generation-fast rendering-instant playback", which significantly reduces the waiting and repeated export steps for users in the subsequent video editing stage, and improves the overall video editing efficiency.

[0064] In this embodiment, during the generation of the description file for the target special effects, the algorithm model uses continuous time variables as its core element. It describes the effect parameter values ​​at any given time by analyzing mathematical equations (such as damped vibration, Bézier curves, or Gaussian convolutions) that are compatible with the target special effects. To reduce computational complexity, a discrete sampling strategy is used to extract points from the continuous equations, obtaining a small number of representative control points. These control points constitute a sparse set of keyframes. Then, using linear or cubic interpolation algorithms, intermediate frames are automatically filled between adjacent keyframes, thereby reconstructing a smooth animation curve and significantly reducing the computational burden of frame-by-frame calculation. The sampling process unfolds according to the target algorithm model, which maps the semantic features of the target special effects style to specific parametric equations. By substituting the step size determined by the frame rate, the parameter values ​​corresponding to each frame number are calculated sequentially, generating a mapping table between frame numbers and parameter values. This mapping table is a sequence of effect keyframes carrying timestamps, with each record precisely indicating the effect intensity or geometric transformation amount to be applied at a specific moment in the base video. Since the equation itself is uniquely determined by the style, the sampling results naturally maintain a high degree of consistency with the target effect, thus ensuring that the visual characteristics of the effect meet the user's expectations. To obtain diverse animation effects under the same style, random perturbations or pseudo-random sequences are introduced within the allowed range of control parameters. Each perturbed parameter set corresponds to a continuous curve with significantly different shapes. After processing with the same sampling rules, multiple keyframe sequences with different visual appearances can be generated. After these sequences are written into the description file, the rendering engine reads them according to the frame number and performs real-time interpolation, resulting in rich variations within the same effect style, meeting the needs of batch comparison. In addition, the interpolation of keyframes does not cause animation breaks because the player has an interpolation mechanism. When parsing the description file, the rendering engine performs automatic interpolation calculations between adjacent keyframes, transforming discrete points into a continuously changing parameter stream. What the user sees is a dense sequence of interpolated frames, resulting in a smooth and coherent visual experience. In other words, only sparse and effective control points are needed to reconstruct the complete animation based on the engine's interpolation capabilities, balancing computational efficiency and visual quality, ultimately achieving an efficient effect preview chain of "generate first, then render on demand".

[0065] Based on this, a preferred implementation method for generating multiple target effect description files by calling a target mathematical algorithm model adapted to the target effect style and the base video from multiple mathematical algorithm models includes: matching algorithm models among multiple mathematical algorithm models based on the target effect style; if no match is found, generating a target mathematical algorithm model based on the target effect style and the base video; introducing a random perturbation sequence or pseudo-random sequence within the allowable range of the control parameters of the target mathematical algorithm model to generate multiple sets of differentiated control parameter values. The allowable ranges for different control parameters may differ. Each numerical value corresponds to a unique motion curve, ensuring the generation of N distinct visual effect description files within the same style, allowing users to compare and select between them in the presentation layer. Differential control parameter values ​​are discretely sampled at a preset frame rate, meaning each group of differential control parameters is sampled separately to obtain a sequence of effect keyframes with timestamps. These keyframe sequences record the effect attribute value at each time point. Based on the effect keyframe sequences corresponding to each group of control parameter values, multiple description files for target effects are generated. For example, encapsulating the effect keyframe sequences corresponding to each group of control parameter values ​​yields multiple description files for target effects. This ensures that the description files are both style-appropriate and diverse, providing ample comparison material for the presentation layer at once, aligning with the overall "batch generation - instant preview" rhythm, reducing subsequent repeated calls, significantly reducing user waiting and repetitive export steps in the subsequent video editing stage, and improving overall video editing efficiency.

[0066] The above embodiments are not limited to the implementation method of using a discrete sampling strategy to perform point sampling operations on the continuous equation to obtain a small number of representative control points. For example, point sampling operations can be performed on the continuous equation at a fixed time step (e.g., every 15 seconds) to obtain a small number of representative control points, which constitute a sparse set of keyframes. Another example is to obtain a small number of representative control points based on physical laws and human visual characteristics. Here, physical laws refer to quantifiable natural or mathematical laws that determine the shape of the animation curve, such as: the damping coefficient ζ and natural frequency ω in the damped vibration equation x(t)=A·e^(-ζωt)·sin(ωt+φ); the cubic polynomial B(t)=ΣC_i·B_{i,3}(t) of the Bézier curve and its curvature extrema; the acceleration-time relationship a(t)=d in kinematics. 2 x / dt 2These equations provide objective extreme points indicating "when changes are fastest and slowest," serving as the mathematical basis for "visually sensitive moments." Human visual characteristics refer to experimentally verified physiological or psychological visual thresholds, which can be directly used to reduce computational load without being noticed. Examples include the critical flicker frequency (CFF): 50–60 Hz, above which the human eye perceives as continuous; the Weber-Fechner law: perceived intensity ΔI is proportional to the relative stimulus amount ΔI / I, used for logarithmic sampling; motion occlusion effect: tiny changes within 20–40 ms near acceleration extremes are most sensitive; just-perceptible difference (JND): when brightness or displacement changes < 1 JND, 90% of observers cannot distinguish them. Therefore, the video processor first uses "physical laws" to find objective extremes or inflection points on the curve, then uses "human visual characteristics" to determine whether these extreme points require encrypted sampling, while sparse sampling is used for other areas. This results in a sharp reduction in computational load while maintaining subjective visual integrity, achieving "visually lossless lightweight sampling" in the patented sense.

[0067] Optionally, based on physical laws and the characteristics of human vision, a small number of representative control points are obtained, including: an adaptive sampling method based on the "critical flicker frequency (CFF)" and a logarithmic sampling method based on the "Weber-Fechner law". The adaptive sampling method based on the "critical flicker frequency (CFF)" is based on the principle that the human eye's perception threshold for flicker is approximately 50–60 Hz, but its perception of motion smoothness is nearly continuous at ≥25 fps. Therefore, calculation is not required for each frame; only dense sampling is needed in visually sensitive regions (such as acceleration extrema and curvature peaks), while sparse sampling is used in other regions. The logarithmic sampling method based on the "Weber-Fechner law" is based on the principle that the human eye's perception of brightness / motion changes follows a logarithmic relationship (ΔI / I ≈ constant). The system uses sparse sampling in low-amplitude regions and dense sampling in high-amplitude change regions to keep the subjective rate of change constant.

[0068] Based on the above principles, an adaptive sampling method based on the "critical flicker frequency (CFF) of the human eye" includes: calculating the second derivative of the target algorithm equation (such as damped vibration) and extracting the acceleration extrema as the "visually sensitive moment"; using a first value (such as 25fps) as a benchmark, inserting N keyframes (such as 2-3) within ±40ms of the sensitive moment; sparsely sampling the remaining areas at a fixed step size (such as 15 seconds); the final keyframe set is approximately 1 / 10 of the original, but the visual error is <1 JND (just perceptible difference). Compared to sampling at equal intervals of 15 seconds, the computational load is reduced by 90%, and the subjective smoothness score (DMOS) is improved by ≥15%. In conjunction with the overall "lightweight first frame + asynchronous high-definition" link, it achieves millisecond-level first frame and imperceptible interpolation.

[0069] The logarithmic sampling method based on the Weber-Fechner law includes: normalizing the target algorithm equation (such as a Bézier curve) to the [0,1] amplitude domain; taking the natural logarithm of the amplitude values ​​to construct a logarithmic time axis; sampling at equal intervals on the logarithmic axis (e.g., every 0.1 log-unit), and then mapping back to linear time. The results show that the sampling interval in the low amplitude region is >30 seconds, and in the high amplitude region it is <5 seconds, reducing the total number of keyframes by 70%. Subjective flicker perception is reduced by 40% (based on ITU-R BT.500 subjective testing), and the peak interpolation error is <0.5%. Combined with the "description file first" strategy, remote preview bandwidth usage is further reduced by 60%, significantly enhancing the cross-platform real-time experience.

[0070] In summary, based on the critical flicker frequency of the human eye or the Weber-Fechner law, adaptive encrypted sampling can be performed on the logarithmic domain of the acceleration extrema or amplitude of the target algorithm equation to generate a sparse but visually lossless set of keyframes, thereby reducing the computational load of discrete sampling and maintaining subjective smoothness.

[0071] The following are exemplary descriptions of how to determine the keyframes for jitter, scaling, and blur effects, respectively.

[0072] Shaking effect - Damped vibration algorithm model:

[0073] Formula for damped vibration curve: Where A is the initial amplitude, and ζ is the damping ratio (range 0.001-0.01). φ is the natural frequency (range 0.1-0.12), t is time (number of frames), and φ is the initial phase difference.

[0074] Implementation logic: FOR i = 0 TO duration STEP 15 / / One keyframe every 15 frames;

[0075] x_offset=intensity*sin(i*0.1)*exp(-i*0.001);

[0076] y_offset=intensity*cos(i*0.12)*exp(-i*0.001);

[0077] rotation=intensity*sin(i*0.08)*0.5;

[0078] The relevant parameters for any keyframe are (frame = i, x = x_offset, y = y_offset, rotation = rotation), which represent the jitter state corresponding to this keyframe.

[0079] Scaling effect - Bézier curve interpolation algorithm model:

[0080] Bézier curve formula: B(t) = (1-t) 2 P0+2(1-t)tP1+t 2 P2; where P0 is the initial scaling value.

[0081] P1 is the control point (dynamically calculated), P2 is the end scaling value, t is the normalization time, and the interval of t is [0,1].

[0082] Implementation logic:

[0083]

[0084] The relevant parameters (scaled rectangle coordinates) of any keyframe are (x, y, width, height), which represent the scaling state corresponding to this keyframe.

[0085] Blur Effect - Gaussian Blur Algorithm Model

[0086] Gaussian curve formula: G(x,y)=(1 / 2πσ 2 )*e^(-(x 2 +y 2 ) / 2σ 2 ); where σ is the fuzzy radius parameter.

[0087] Implementation logic:

[0088]

[0089] END IF

[0090] The relevant parameters for any keyframe are (frame, intensity_current, angle), which represent the blurred state of this keyframe.

[0091] In one optional embodiment, a target mathematical algorithm model is generated based on the target special effects style and the base video, including: selecting a base algorithm model from the base algorithm template library that matches the curve change trajectory represented by the target special effects style as the target base algorithm template; optimizing the parameters of the target base algorithm template based on the sample set of the target special effects style to obtain a target algorithm model that matches the target special effects style, wherein each sample in the sample set contains an example pair consisting of an original material frame sequence and a target special effects frame sequence.

[0092] When selecting a basic algorithm model from the basic algorithm template library that matches the curve change trajectory representing the target special effects style, one basic algorithm model may not be compatible with the target special effects style and the basic video. In this case, in addition to selecting one basic algorithm model, two or more basic algorithm models can be selected to synthesize the target basic algorithm template based on two or more basic algorithm models. Based on this, a basic algorithm model that matches the curve change trajectory representing the target special effects style is selected from the basic algorithm template library as the target basic algorithm template. Optional implementation methods include: dividing the expected trajectory of the target special effects style into several sub-intervals on the time axis; calculating the shape similarity between each sub-interval and the curve segment of each basic algorithm model (using Dynamic Time Warping (DTW) or piecewise Bézier control point distance); sorting by similarity from high to low; selecting the Top-N models that are geometrically complementary (e.g., one segment responsible for "ease in," one for "oscillation," and one for "ease out") as candidates; then checking the compatibility of these candidate models with the basic video metadata in terms of frame rate and amplitude range; filtering out models that exceed the allowed range; and finally retaining two or more complementary and compatible basic algorithm models for subsequent splicing or weighted synthesis to ensure that the synthesized target basic algorithm template has the smallest error with the expected curve on the entire timeline and that the parameters are continuously differentiable.

[0093] In another optional embodiment, a target mathematical algorithm model is generated based on the target special effects style and the base video, including: acquiring motion effect description information in natural language form; inputting the motion effect description information and / or the target special effects style into a dynamic algorithm generation model; mapping curve control points based on key intent parameters of the motion effect description information and / or the target special effects style to obtain a mapping result; generating an executable object of the corresponding algorithm model based on the mapping result, which serves as the target algorithm model adapted to the motion effect description information and / or the target special effects style. The executable object refers to... The key intent parameters include at least: change trend, change range, and change period length. The trend of change refers to the macroscopic direction of the curve slope over time, such as linear, slow-in, slow-out, elastic, and bounce, which is used to qualitatively describe the speed and rhythm of motion effects such as "slow at first and then fast" or "fast at first and then slow". The range of change refers to the upper and lower limits of the values ​​that the special effect parameters can take, such as displacement 0-20px and scaling 1.0-1.2, which is used to limit the extreme values ​​of the vertical axis of the curve to prevent overshoot or visual distortion. The length of the change cycle refers to the time or number of frames required to complete one complete fluctuation, which determines the span of the horizontal axis of the curve and is directly related to the speed and repetition frequency of the animation.

[0094] The above implementation method, based on the target special effects style and the basic video generation target mathematical algorithm model, provides two pathways: "sample set regression" and "large model semantic mapping." "Sample set regression" uses historical frame pairs as supervision, prompting template parameters to approach the optimal state of visual effect; "large model semantic mapping" directly converts natural language into curve control points without manual parameter adjustment. These two modes share a single output interface, allowing the video processor to both inherit experience and quickly adapt to new vocabulary, significantly improving the model's hit rate and parameter accuracy. This reduces duplicate rendering caused by algorithm mismatches, improving rendering efficiency. Users can preview the special effects parameter video in a timely manner and quickly select the special effects template, thereby improving the efficiency of subsequent video editing based on the special effects template. In other words, by splitting the desired trajectory into sub-intervals and calculating the DTW / control point distance, the video processor can accurately match multiple complementary curves such as "ease-in-oscillation-ease-out" on the time axis. After filtering by amplitude range and period length, it retains two or more basic algorithm models that are both stylistically consistent and compatible with the basic video frame rate and resolution. Subsequent stitching or weighted synthesis yields a target template with the smallest error throughout and continuous differentiability, significantly improving model hit rate and parameter accuracy. Subsequently, both sample set regression and large model semantic mapping are fine-tuned based on this synthesized template, making the final generated keyframe sequence naturally fit the user's language description, reducing repeated rendering caused by single template mismatch, and thus working in conjunction with the overall "accurate generation-fast rendering-instant playback" link to further shorten the special effects confirmation time and reduce computational idle time.

[0095] Optionally, the motion effect description information and / or target effect style are input into the dynamic algorithm generation model, and curve control points are mapped based on the key intent parameters described by the motion effect description information and / or target effect style to obtain the mapping result. This includes: inputting the motion effect description information and / or target effect style into the dynamic algorithm generation model, performing semantic extraction on the dynamic description information and / or target effect style to obtain at least one motion effect intent semantic unit, whereby the motion effect intent semantic unit is a discrete semantic unit used to represent the user's expected motion effect category and its changing trend; mapping at least one motion effect intent semantic unit to a set of key intent parameters based on the intent and algorithm parsing mapping library, which serves as the set of control parameters required by the corresponding algorithm model; and performing at least amplitude scaling and period scaling processing on each initial control point in the initial control point template based on the constraint rules of the control parameter set to obtain a target control point array, which serves as the mapping result. The constraints on the control parameter set should include at least the following aspects: amplitude boundaries: the ordinate values ​​of the scaled control points must not exceed the maximum amplitude A_max and minimum amplitude A_min allowed by the model; period boundaries: the stretching factor of the abscissa (time axis) must not cause the total curve duration to exceed the basic video duration T_video, nor be less than the minimum period T_min; monotonicity preservation: the slope signs between adjacent control points after scaling must not be reversed to ensure that the curve trend (gradual in / gradual out / oscillation) matches the user's intention. Figure 1 For continuity, the distance between adjacent control points must not be less than the minimum resolution Δt to avoid excessively dense inflection points causing interpolation oscillations. For semantic consistency, the initial weights of the verb-adjective-numerical triplet mapping must not change more than the preset tolerance ε before and after scaling to prevent semantic drift. The ambiguous text is decomposed into discrete "type-trend-amplitude" labels using motion effect intention semantic units, then mapped to a set of key intention parameters. Finally, the target array is obtained by scaling the initial control points only by amplitude scaling and periodic stretching. This process is entirely linear transformation, with low computational complexity, yet it accurately places a user's sentence into a numerical range, ensuring that subsequent discrete sampling results are consistent with user expectations, improving initial preview satisfaction, preview efficiency, and the efficiency of special effects template selection, thereby improving the user's subsequent video editing efficiency.

[0096] The intent and algorithm parsing mapping library includes, but is not limited to: a verb and intent sub-library, an adjective and trend sub-library, and a numerical and range sub-library. An optional implementation method based on the intent and algorithm parsing mapping library to map at least one motion effect intent semantic unit to a set of key intent parameters includes: identifying the word attributes represented by at least one motion effect intent semantic unit, where word attributes include verbs, adjectives, and numerical nouns; mapping the verb attributes represented by at least one motion effect intent semantic unit to effect style types based on the verb and intent sub-library; mapping the adjective attributes represented by at least one motion effect intent semantic unit to change trends based on the adjective and trend sub-library; and mapping the numerical noun attributes represented by at least one motion effect intent semantic unit to change magnitude or change period based on the numerical and range sub-library.

[0097] The above embodiments, through "semantic unit-sub-library mapping," instantly transform natural language into machine-readable four-dimensional parameters of "style-trend-amplitude-period," significantly reducing the user input threshold: the verb sub-library locks in the effect category, the adjective sub-library determines the curve slope change, and the numerical sub-library quantifies the amplitude and period. The sum of these three can obtain the key intent parameter set in milliseconds. Subsequently, only amplitude scaling and period scaling are performed on the initial control point template, without the need to reconstruct the equation, ensuring that the curve shape is consistent with the user's language sense and with extremely low computational overhead. The target control point array generated in this way can produce a high-fidelity keyframe sequence through discrete sampling, which can be directly written into the description file and seamlessly connected with the subsequent "on-demand rendering" link; the accuracy of semantic mapping is improved, reducing repeated rendering caused by parameter misunderstanding, thereby strengthening the overall efficient closed loop of "accurate generation-fast rendering-instant playback," further shortening the effect confirmation time in video editing, and thus helping users' video editing efficiency.

[0098] Optionally, based on the constraint rules of the control parameter set, at least amplitude scaling and period scaling are performed on each initial control point in the initial control point template to obtain the target control point array. This includes: first, parsing the control parameter set to obtain the upper and lower limits of amplitude and the period boundary, and calculating the horizontal and vertical scaling factors accordingly; then, multiplying the horizontal coordinate of the initial control point template by the period scaling factor and the vertical coordinate by the amplitude scaling factor to complete a linear affine transformation; immediately after the transformation, boundary bounce is performed, that is, coordinates exceeding the allowable interval are truncated to the nearest boundary value, while keeping the slope sign of adjacent points unchanged to ensure that the curve trend does not reverse; if there are too dense control points, they are redistributed equally at the minimum time step and the corresponding vertical coordinates are updated synchronously to maintain continuous differentiability; finally, semantic consistency is checked on the parameters before and after scaling. If the change in mapping weight exceeds the tolerance, a small offset is applied within the boundary to correct it. The final output is a target control point array that satisfies both the user's intent and the physical constraints of the model, which can be directly used for discrete sampling to generate keyframe sequences. Through a five-step closed loop of "analysis-scaling-boundary bounce-trend preservation-semantic correction," the user's colloquial intentions regarding "amplitude and period" are instantly transformed into mathematically rigorous control point coordinates. This ensures that the scaled curve remains within the allowable range while maintaining its original trend, avoiding visual jumps caused by exceeding limits or slope reversals. Boundary truncation and equidistant redistribution reduce overly dense inflection points, lowering the computational load and interpolation errors of subsequent discrete sampling. This allows for the output of high-fidelity keyframe sequences during the description file generation stage. Consequently, the algorithm engine layer can deliver usable parameters in one go without repeated trial and error, seamlessly connecting with the "on-demand rendering" link of the rendering adaptation layer, shortening the overall time from language input to first frame preview. Simultaneously, consistent parameter trends reduce repeated rendering caused by model mismatch, further strengthening the efficient closed loop of "accurate generation-fast rendering-instant playback." This improves preview efficiency and the efficiency of special effects template selection, thereby enhancing subsequent video editing efficiency.

[0099] Optionally, an executable object of the corresponding algorithm model is generated based on the mapping result, serving as the target algorithm model adapted to the motion effect description information and / or the target effect style. This includes: injecting the mapping result into a basic algorithm template corresponding to the motion effect description information and / or the target effect style to form a parameterized algorithm equation; discretely sampling the parameterized algorithm equation at a preset frame rate to generate a keyframe parameter sequence; and performing a compilation operation based on the keyframe sequence to obtain an executable object of the parameterized equation, serving as the target algorithm model adapted to the motion effect description information and / or the target effect style. By injecting the mapped control points into the basic template, a complete parameterized equation is formed. The keyframe parameter sequence is obtained by sampling at a preset frame rate, and then a native evaluation function is generated through lightweight JIT compilation. This executable object can be directly loaded by the rendering engine, and only array index lookup is performed at runtime, without the need to repeatedly parse the description file or recalculate the curve. The first frame rendering latency is reduced to the millisecond level. Combined with the "on-demand rendering" strategy, this significantly shortens the user's waiting time, thereby improving the video editing efficiency of subsequent users.

[0100] In one optional embodiment, multiple target effect description files and base video description files are merged to obtain multiple effect reference video description files. This includes: obtaining a base video frame sequence and timestamps for each frame; obtaining multiple effect keyframe sequences contained in each target effect description file; inserting each effect keyframe sequence into the corresponding base video frame according to its timestamp to form a composite frame sequence with effect attributes; and generating multiple effect reference video description files based on each composite frame sequence. By inserting the effect keyframe sequence into the attribute nodes of the corresponding frames of the base video only according to the timestamp, without copying the original pixel data, the size of the base video description file remains constant. A single insertion operation can generate multiple composite frame sequences, allowing subsequent rendering layers to complete image compositing simply by reading the newly added attributes. This reduces input / output operations and memory usage, providing lightweight input for rapid preview and thus improving the efficiency of subsequent video editing for users.

[0101] Optionally, multiple target effect description files and the description file of the base video can be merged to obtain multiple effect reference video description files. This includes: obtaining the description file of the base video, which at least contains base video tracks, material references, and timeline metadata; obtaining multiple target effect description files, each containing an effect track and its keyframe attribute nodes; concatenating the effect track nodes to the base video description file at the timeline level according to a time reference to form a merged timeline structure; adding track blending mode attributes to the merged timeline structure and generating multiple effect reference video description files. This fusion implementation only performs the "track concatenation + blending mode attribute injection" operation at the timeline level, without requiring copying or re-encoding of the base video frame data, thus enabling multiple effect tracks to be mounted to the same base video description file at once, thereby generating multiple effect reference video description files. This implementation significantly reduces storage and transmission overhead. For example, the description file size is kept at the KB level, and only a lightweight XML / MLT needs to be sent during remote calls, which aligns with the overall "description-first, on-demand rendering" strategy and reduces network bandwidth usage. The rendering adaptation layer can directly read the merged timeline structure and parse the effect tracks in parallel according to the frame number, achieving multi-task concurrency in a single engine, thereby shortening the total time for batch previews. The track blending mode attribute allows effects and base video to be synthesized in real time at the pixel level according to specified algorithms (overlay, color filtering, alpha blending, etc.), ensuring visual consistency while providing standardized input for subsequent "second-level first frame" and "asynchronous high-definition completion," further strengthening the efficient closed loop of "accurate generation - fast rendering - instant playback," and comprehensively improving video editing efficiency.

[0102] In this embodiment, each preview position in the display area also includes a preview control; responding to preview operations for a target preview position includes: when the hover time on the target display position exceeds a preset time threshold, the corresponding hover action is considered a preview operation; or, when a trigger action is generated based on the preview control on the target preview position, the trigger action is considered a preview operation. Setting two trigger dimensions, hover timeout and click, can reduce misplays by leveraging the natural operation of the mouse and is also compatible with scenarios where there is no hover effect on mobile devices; binding the trigger signal directly to the rendering service request ensures seamless connection between the generation of the description file and the playback link, thereby improving the responsiveness of the interaction and the smoothness of the preview.

[0103] In this embodiment, each preview position in the display area also includes a details control. In response to a trigger operation on the details control of the target preview position, a details page is displayed. The details page includes a previewable target effect reference video, index information of the description file of the target effect reference video, and description information contained in the description file. In response to a modification operation on the description information, the modified description information is obtained, and the modified description information is updated to the description file based on the index information. An exemplary schematic diagram of the details page is shown below. Figure 2b As shown, the details control constructs a "click-and-review" closed-loop mode within the preview area. After clicking, the overlay will present the target effect reference video, description file index, and all parameters at once. Users can edit while watching without jumping to another page; the modified description information is immediately written back to the original file through the index. The video processor does not need to regenerate the model, but only incrementally updates the keyframe values ​​or attribute fields, thereby significantly shortening the parameter adjustment and preview cycle. This on-site modification mechanism is coupled with the overall "description-first, on-demand rendering" chain. After the write-back is completed, the rendering adaptation layer can immediately use the latest description file to carry out the next round of lightweight first-frame rendering, achieving the effect of "view immediately after modification"; at the same time, it can avoid repeated exporting due to parameter errors, reduce the phenomenon of idle computing power, further strengthen the efficient closed-loop mode of "accurate generation-fast rendering-instant review", and comprehensively improve the efficiency of video editing.

[0104] In this embodiment, multiple rendering engines have priorities, which are determined based on at least one of the following indicators: performance, resource utilization, historical success rate, task matching degree, network location, and real-time health status. Responding to a rendering service request, selecting a target rendering engine from among the multiple rendering engines includes: responding to the rendering service request, determining the priority of each rendering engine based on at least one indicator to obtain a priority queue; performing pre-checks on the rendering engines in the priority queue in order of priority based on a rendering engine degradation mechanism; if the current priority rendering engine meets the pre-check conditions, then the current priority rendering engine is selected as the target rendering engine; if the current priority rendering engine does not meet the pre-check conditions, it is automatically downgraded to the next priority rendering engine, and pre-checks the next priority rendering engine until a rendering engine that meets the pre-check conditions is determined; if all priority rendering engines do not meet the pre-check conditions, then a placeholder video is obtained and displayed on the target display position. The rendering engine degradation mechanism is roughly as follows: a priority queue, dynamically sorted by real-time metrics, with high-priority engines receiving pre-checks first; pre-check scripts, employing dual commands for rendering function testing and effects parameter identification verification, marking a downgrade if either fails; automatic degradation, where a high-priority engine fails and slides to the next best, cycling until a healthy engine is found; and final placeholder, where all engines are unavailable and a pre-generated stylized video is switched to ensure 100% playability of the preview slot. This three-level filtering mechanism—multi-metric priority queue, pre-check, and automatic degradation—transforms single-point failures into seamless switching. Upon arrival of a rendering service request, candidate engines are dynamically scored and a priority queue is generated based on performance, resource utilization, historical success rate, task matching, network location, and real-time health status. Pre-checks, consisting of rendering test commands and effects parameter identification commands, are executed sequentially, instantly eliminating unhealthy nodes. High-priority engines that fail are automatically downgraded to the next best, ensuring that rendering tasks always have runnable instances. If all engines fail, a placeholder video consistent with the target effect style is immediately displayed, ensuring the preview screen remains smooth and lag-free. This negative degradation strategy is directly coupled with the front-end's "second-level first frame" requirement, allowing users to receive visual feedback instantly upon hovering or clicking, without waiting for fault recovery. Simultaneously, it avoids invalid tasks consuming cluster resources, enabling the rapid processing of numerous concurrent preview requests after batch description file generation. This strengthens the efficient closed loop of "accurate generation, rapid rendering, and instant playback," significantly shortening the effect confirmation time during video editing and improving the efficiency of subsequent video editing for users.

[0105] Optionally, in response to a rendering service request, the priority of each rendering engine is determined based on at least one metric, including: performing mathematical calculations based on at least one of the following metrics: performance, resource utilization, historical success rate, task matching degree, network location, and real-time health status of each rendering engine; and determining the priority of each rendering engine based on the results of the mathematical calculations. The mathematical calculations include, but are not limited to, weighted summation and variance calculation. Therefore, calculating the priority of each rendering engine using multiple metrics can improve the accuracy of prioritization, laying the foundation for improving the user's video editing efficiency in the future.

[0106] Optionally, the rendering engines in the priority queue are pre-checked sequentially according to priority. This includes: calling a preset rendering test command to attempt to run the rendering function of the executable file and outputting the first feedback information; calling a preset special effects test command to attempt to run the special effects parameter recognition function of the executable file and outputting the second feedback information; if both feedback information indicates that the test is successful, the current priority rendering engine is determined to meet the pre-check conditions; if at least one feedback information indicates that the test fails, the current priority rendering engine is determined to not meet the pre-check conditions. Through the two scripts, the rendering test command and the special effects parameter recognition command, dual verification of the executable file and special effects support is completed before the task is officially issued. When both the first and second feedback information indicate that the engine is in good condition, the timing of fault detection can be advanced from "mid-rendering" to "the moment of scheduling." This verification result directly affects the priority queue. For failed nodes, their weight is immediately reduced or they are removed, and subsequent requests will automatically bypass these nodes. This avoids a large number of preview tasks falling into invalid scheduling after batch description file generation, ensuring that the overall "quasi-generation-fast rendering" link is not blocked, shortening user waiting time, and thus improving the video editing efficiency of subsequent users.

[0107] Optionally, the rendering adaptation layer is also used to monitor for failure events of any priority rendering engine during the rendering process. When any failure event is detected, the rendering function of that priority rendering engine is shut down, and another priority rendering engine that meets the pre-check conditions is selected according to the rendering engine degradation mechanism. The rendered data of any priority rendering engine is converted into a data format supported by the other priority rendering engine, and the converted rendered data is migrated to the other priority rendering engine so that the other priority rendering engine can continue to render the description file of the target effect reference video. By capturing engine crash signals in real time through heartbeat or return codes through the rendering adaptation layer, the faulty process is shut down on the spot and the degradation mechanism is triggered. The rendered frames are immediately converted into a format compatible with the lower-level engine and then migrated for continued rendering, achieving a seamless switching with "no black screen and no loss of progress". This self-healing capability, combined with the second-level first frame and asynchronous high-definition completion process, ensures that remote users still receive continuous visual feedback in scenarios with weak networks or node failures. At the same time, rendered data is reused instead of recalculated, significantly reducing cluster duplication overhead and further compressing the total time for batch concurrent previews, thus strengthening the overall efficient closed loop.

[0108] Optionally, the video processor also includes a file management layer. This layer stores description files for reference videos of various special effects styles, each with its own storage address. Furthermore, it deletes the description file of any reference video when it exceeds a preset storage period. By automatically categorizing by style-date hierarchy and setting storage periods, expired description files and their preview videos are periodically deleted, freeing up storage space and preventing index bloat and retrieval delays caused by massive accumulation of lightweight XML / MP4 files. After space reclamation, newly generated description files can continue to be written at high speed, aligning with the "description-first, on-demand rendering" strategy. This ensures that the remote preview bit always points to a valid address, preventing interruptions to the creation process due to disk exhaustion and maintaining long-term stability in video editing efficiency.

[0109] Optionally, users can download the corresponding effect template from the details page. The video processor also includes an editing layer where users can select an effect template and, based on the selected template and the target video, generate the desired large-scale effect video with a single click.

[0110] At this point, the execution steps of each layer of the video processor have been completed, significantly reducing the waiting and repetitive exporting steps for users in the subsequent video editing stage, and improving the overall video editing efficiency.

[0111] For ease of understanding, the above technical solutions will be described in general below.

[0112] Step 1: Interface Initialization

[0113] The platform presentation layer loads Bootstrap 5 and Font-Awesome resources before the terminal, then starts the Vue / React instance and establishes a global state repository; after obtaining the list of effects styles through the RESTful interface GET / api / styles, it renders the three-column layout of sidebar-main content area-operation panel according to the screen width and responsive rules, and sets the first style as the default selection, thus completing the initialization of the effects configuration interface.

[0114] Step 2: Configure Special Effects Styles and Generate Display Space

[0115] In response to the user's selection of the special effects style configuration option, the algorithm engine layer immediately obtains the target special effects style and calls the target mathematical algorithm model that is compatible with the style and the base video from multiple mathematical algorithm models to generate multiple description files for the target special effects; then these description files are merged with the description file of the base video to obtain description files for multiple special effects reference videos corresponding to each preview position; the display layer synchronously presents the preview positions in the special effects video preview area, and each position is associated with the index and preview control of the corresponding description file.

[0116] Step 3: Preview Request and Rendering Engine Selection

[0117] The rendering layer responds to the user's trigger operation on the preview control in the target preview position by sending a rendering service request to the rendering adaptation layer. Based on the request, the rendering adaptation layer performs pre-examination on multiple rendering engines in a priority queue. If the current priority rendering engine meets the pre-examination conditions, it is determined as the target rendering engine and its identification information is returned. Otherwise, it is automatically downgraded to the next priority engine until an available engine is selected.

[0118] Step 4: Special Effects Reference Video Rendering and Display

[0119] The rendering layer calls the engine based on the returned target rendering engine identification information, performs rendering according to the description file of the target effect reference video, and renders the generated effect reference video to the corresponding preview position, allowing the user to decide whether to use the target effect for subsequent video editing based on the preview effect; if all rendering engines are unavailable, a placeholder video is enabled and presented in the preview position to ensure that the preview link is not interrupted.

[0120] Step 5: Batch generation and result output

[0121] Once the user inputs the quantity and confirms it through the batch operation control, the algorithm engine layer repeats the logic of step 2, batch outputting description files for differentiated special effects reference videos; the file management layer stores them according to the "style / date" structure and writes the storage address of each file into the index; after the task is completed, the display layer refreshes the special effects reference video display area, and the user can trigger rendering and preview on the new preview position, realizing a closed loop of batch generation - real-time rendering - instant preview.

[0122] Figure 3 This is a flowchart illustrating a video editing method provided as an exemplary embodiment of this application. Figure 3 As shown, the method includes:

[0123] 301. Display the special effects configuration interface, which includes special effects style configuration items and a special effects reference video display area. The display area includes multiple preview slots for displaying special effects reference videos.

[0124] 302. Respond to the configuration operation through the special effects style configuration item, obtain the target special effects style; from multiple mathematical algorithm models, call the target mathematical algorithm model that is compatible with the target special effects style and the base video, and generate multiple description files of target special effects; merge the description files of multiple target special effects and the description file of the base video to obtain multiple description files of special effects reference videos, and each description file of special effects reference video corresponds to a preview bit;

[0125] 303. Respond to the preview operation for the target preview position and generate a rendering service request. The target preview position corresponds to the description file of the target effect reference video.

[0126] 304. In response to a rendering service request, select a target rendering engine from multiple rendering engines, wherein the target rendering engine has identification information;

[0127] 305. Based on the target rendering engine's identification information, call the target rendering engine and render the target effect reference video to the target preview position according to the description file of the target effect reference video, so that the user can determine whether to select the target effect in the target effect reference video for subsequent video editing based on the preview effect.

[0128] For detailed implementation methods of each step, please refer to the description in the foregoing video processor embodiments, which will not be repeated here.

[0129] Figure 4 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this application. For example... Figure 4 As shown, the electronic device includes: a memory 40a and a processor 40b;

[0130] The memory 40a is used to store computer programs; the processor 40b is coupled to the memory 40a and is used to execute the computer programs for:

[0131] Furthermore, such as Figure 4 As shown, the server device also includes other components such as a communication component 40c, a power supply component 40d, an audio component 40e, and a display screen 40f. Figure 4 The diagram only shows some components and does not mean that the computer device includes only these components. Figure 4 The components shown.

[0132] An exemplary embodiment of this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the above method, which will not be described in detail here.

[0133] An exemplary embodiment of this application also provides a computer program product, including: a computer program that, when executed by a processor of an electronic device, causes the processor to perform the steps in the above method, which will not be described in detail here.

[0134] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0135] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0139] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0142] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A video processor, characterized in that, include: The system comprises a presentation layer, an algorithm engine layer, a rendering adaptation layer, and a rendering layer, wherein the algorithm engine layer is decoupled from the rendering layer. The display layer is used to display the special effects configuration interface, which includes special effects style configuration items and a special effects reference video display area. The display area includes multiple preview slots for displaying special effects reference videos. The algorithm engine layer is used to respond to configuration operations through the special effects style configuration items, obtain the target special effects style, and call the target mathematical algorithm model that is adapted to the target special effects style and the basic video from multiple mathematical algorithm models to generate multiple description files of target special effects. The description files of the multiple target special effects and the description file of the base video are merged to obtain the description files of multiple special effect reference videos, and each description file of the special effect reference video corresponds to a preview bit; The rendering layer is used to respond to preview operations for the target preview position and send a rendering service request to the rendering adaptation layer. The target preview position corresponds to the description file of the target special effects reference video. The rendering adaptation layer is used to respond to the rendering service request, select a target rendering engine from multiple rendering engines, and return the identification information of the target rendering engine to the rendering layer; The rendering layer is further configured to invoke the target rendering engine based on the identification information of the target rendering engine, and render the target effect reference video to the target preview position according to the description file of the target effect reference video, so that the user can determine whether to select the target effect in the target effect reference video for subsequent video editing based on the preview effect.

2. The video processor according to claim 1, characterized in that, From multiple mathematical algorithm models, a target mathematical algorithm model adapted to the target effect style and base video is called to generate multiple description files for the target effects, including: Based on the target special effects style, algorithm model matching is performed among multiple mathematical algorithm models; If no match is found, a target mathematical algorithm model is generated based on the target special effects style and the base video. Within the allowable interval of the control parameters of the target mathematical algorithm model, random perturbation sequences or pseudo-random sequences are introduced to generate multiple sets of differentiated control parameter values; The differential control parameter values ​​of each group are discretely sampled at a preset frame rate to obtain a special effects key frame sequence with timestamps. The special effects key frame sequence records the special effects attribute value at each time point. Based on the keyframe sequences of special effects corresponding to the differentiated control parameter values ​​of each group, description files for multiple target special effects are generated.

3. The video processor according to claim 2, characterized in that, Based on the target special effects style and the base video, a target mathematical algorithm model is generated, including: A basic algorithm model that matches the curve change trajectory represented by the target special effects style is selected from the basic algorithm template library as the target basic algorithm template; the parameters of the target basic algorithm template are optimized based on the sample set of the target special effects style to obtain a target algorithm model that matches the target special effects style. Each sample in the sample set contains an example pair consisting of an original material frame sequence and a target special effects frame sequence. or, Obtain motion effect description information in natural language form, input the motion effect description information and / or the target special effect style into a dynamic algorithm to generate a large model, perform curve control point mapping based on the key intent parameters of the motion effect description information and / or the target special effect style, and obtain the mapping result; generate an executable object of the corresponding algorithm model based on the mapping result, which serves as the target algorithm model adapted to the motion effect description information and / or the target special effect style; wherein, the key intent parameters include at least: change trend, change range, and change period length.

4. The video processor according to claim 3, characterized in that, The motion effect description information and / or the target effect style are input into a dynamic algorithm to generate a large model. Based on the key intent parameters of the motion effect description information and / or the target effect style, curve control points are mapped to obtain the mapping result, including: The motion effect description information and / or the target special effect style are input into the dynamic algorithm to generate a large model. Semantic extraction is performed on the motion effect description information and / or the target special effect style to obtain at least one motion effect intention semantic unit. The motion effect intention semantic unit is a discrete semantic unit used to characterize the user's expected motion effect category and its changing trend. Based on the intent and algorithm parsing mapping library, the at least one motion effect intent semantic unit is mapped to a set of key intent parameters, which serves as the set of control parameters required by the corresponding algorithm model. Based on the constraint rules of the control parameter set, at least amplitude scaling and period scaling are performed on each initial control point in the initial control point template to obtain a target control point array, which serves as the mapping result.

5. The video processor according to claim 3, characterized in that, An executable object for the corresponding algorithm model is generated based on the mapping result, serving as the target algorithm model adapted to the motion effect description information and / or the target special effects style, including: The mapping result is injected into the basic algorithm template corresponding to the motion effect description information and / or the target special effect style to form a parameterized algorithm equation; The parameterized algorithm equation is discretely sampled at a preset frame rate to generate a key frame parameter sequence. Based on the keyframe parameter sequence, a compilation operation is performed to obtain the executable object of the parameterized equation, which serves as the target algorithm model adapted to the motion effect description information and / or the target special effects style.

6. The video processor according to claim 1, characterized in that, The description files of the multiple target special effects and the description file of the base video are merged to obtain description files of multiple special effects reference videos, including: Obtain the basic video frame sequence and the timestamp of each frame; Obtain multiple keyframe sequences of special effects contained in the special effects description files of each target; Each keyframe sequence of special effects is inserted into the corresponding base video frame according to its timestamp to form a composite frame sequence with special effects attributes; Based on each composite frame sequence, description files for multiple special effects reference videos are generated.

7. The video processor according to claim 1, characterized in that, Each preview position in the display area also includes a preview control; responding to a preview operation for a target preview position includes: When the hovering time on the target preview position exceeds the preset time threshold, the corresponding hovering action is considered a preview operation; or When a trigger action is generated based on the preview control on the target preview position, the trigger action is considered as a preview operation.

8. The video processor according to claim 1, characterized in that, Each preview slot in the display area also includes details controls; and also includes: In response to a trigger operation on the details control on the target preview position, a details page is displayed. The details page includes a previewable target effect reference video, index information of the description file of the target effect reference video, and description information contained in the description file. In response to a modification operation on the description information, the modified description information is obtained, and the modified description information is updated to the description file based on the index information.

9. The video processor according to claim 1, characterized in that, There is a priority among multiple rendering engines, and the priority among the multiple rendering engines is determined based on at least one of the following indicators: performance, resource utilization, historical success rate, task matching degree, network location, and real-time health status of each rendering engine. In response to the rendering service request, selecting a target rendering engine from multiple rendering engines includes: In response to the rendering service request, the priority of each rendering engine is determined based on the at least one metric to obtain a priority queue; Based on the rendering engine degradation mechanism, the rendering engines in the priority queue are pre-checked in order of priority. If the current priority rendering engine meets the pre-detection conditions, then the current priority rendering engine will be used as the target rendering engine. If the current priority rendering engine does not meet the pre-check conditions, it will be automatically downgraded to the next priority rendering engine, and a pre-check will be performed on the next priority rendering engine until a rendering engine that meets the pre-check conditions is determined. If none of the priority rendering engines meet the preflight criteria, a placeholder video is obtained and displayed on the target preview screen.

10. The video processor according to claim 9, characterized in that, The rendering engines in the priority queue are pre-checked sequentially according to their priority, including: Invoke the preset rendering test command, attempt to run the rendering function of the executable file of the current priority rendering engine, and output the first feedback information; Invoke the preset special effects test command, attempt to run the special effects parameter recognition function of the executable file of the current priority rendering engine, and output the second feedback information; If all feedback messages indicate that the test was successful, then the current priority rendering engine is determined to meet the pre-detection conditions. If at least one feedback message indicates that the test has failed, then the current priority rendering engine is determined not to meet the preflight conditions.

11. The video processor according to any one of claims 1-10, characterized in that, The rendering adaptation layer is also used to monitor the failure events of any priority rendering engine during the rendering process of any priority rendering engine. When any failure event is detected, the rendering function of any priority rendering engine is shut down, and other priority rendering engines that meet the pre-detection conditions are reselected according to the rendering engine degradation mechanism. The rendered data of any one priority rendering engine is converted into a data format supported by the other priority rendering engines, and the converted rendered data is migrated to the other priority rendering engines so that the other priority rendering engines can continue to render the description file of the target effect reference video.

12. The video processor according to any one of claims 1-10, characterized in that, The video processor also includes a file management layer. The file management layer is used to store description files of special effects reference videos under each special effects style based on the classification of each special effects style. Each description file has its own storage address. Additionally, if the description file of any special effects reference video exceeds a preset storage period, the description file of that special effects reference video shall be deleted.

13. A video processing method, characterized in that, include: The interface for displaying special effects configuration includes special effects style configuration items and a special effects reference video display area, which includes multiple preview slots for displaying special effects reference videos. In response to the configuration operation of the special effects style configuration item, the target special effects style is obtained; from multiple mathematical algorithm models, the target mathematical algorithm model that is compatible with the target special effects style and the basic video is called to generate multiple description files of target special effects; The description files of the multiple target special effects and the description file of the base video are merged to obtain the description files of multiple special effect reference videos, and each description file of the special effect reference video corresponds to a preview bit; In response to a preview operation targeting a specific preview position, a rendering service request is generated, wherein the target preview position corresponds to a description file of the target effect reference video; In response to the rendering service request, a target rendering engine is selected from multiple rendering engines, and the target rendering engine has identification information; Based on the identification information of the target rendering engine, the target rendering engine is invoked, and the target effect reference video is rendered to the target preview position according to the description file of the target effect reference video, so that the user can determine whether to select the target effect in the target effect reference video for subsequent video editing based on the preview effect.

14. An electronic device, characterized in that, The electronic device includes: a memory and a processor; the memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps of the method of claim 13.

15. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it causes the processor to perform the steps of the method of claim 13.

16. A computer program product, characterized in that, include: A computer program, when executed by a processor of an electronic device, causes the processor to perform the steps of the method as described in claim 13.

Images