Creative video generation method, apparatus, device, storage medium, and program product

Abstract

This application discloses a creative video generation method, apparatus, device, storage medium, and program product. The method includes: acquiring user creative requirement information; collecting materials based on the creative requirement information to obtain a set of materials to be processed; determining at least one creative topic reference material and its event context based on the set of materials to be processed; semantically matching the creative topic reference material with pre-set paradigm features in a paradigm library for each creative topic reference material, and selecting high-quality theme materials whose matching degree meets preset conditions from each creative topic reference material; generating creative topics by calling a pre-trained creative generation model based on the high-quality theme materials and their corresponding event context, and based on the creative topics, generating planning documents, storyboard documents, and finished videos sequentially under the constraints of creative paradigms to obtain a video file containing the target content. This application improves the quality and stability of video generation.

Description

Technical Field

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

[0002] With the development of AI-generated content technology, large-scale models have been gradually applied to various media content production scenarios, such as short videos, cultural and tourism promotions, television programs, and educational courses.

[0003] In related technologies, multi-agent systems are generally regarded as "task parallel splitters" or "integrators of generative models." Multiple agents only pass coarse-grained result files to each other, lacking standardized state management protocols and anomaly recovery mechanisms. Quantitative scoring methods often rely on fixed expert weights, which cannot adapt to the dynamic differences of different content types (such as long videos versus short videos, knowledge-based versus emotion-based content). Multiple agents usually only pass text results to each other, lacking a unified generative state inheritance structure. This results in subsequent generation stages being unable to inherit the previous generation state, leading to problems such as theme deviation, emotional discontinuity, and conflicting camera rhythms. Summary of the Invention

[0004] The main objective of this application is to provide a creative video generation method, apparatus, device, storage medium, and program product, aiming to solve the technical problem of unstable quality in creative video generation in related technologies.

[0005] Firstly, to achieve the above objectives, this application provides a creative video generation method, the method comprising: In response to the user's creative needs, the following steps are executed through multiple intelligent agents: Obtain users' creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed; Based on the set of materials to be processed, identify at least one creative topic reference material and the event context of the material. For each creative topic reference material, semantic matching is performed between the creative topic reference material and the pre-set paradigm features in the paradigm library, and high-quality topic materials that meet the preset conditions are selected from each creative topic reference material. Based on high-quality thematic materials and their corresponding event context, a pre-trained creative generation model is invoked to generate creative topics. Based on the creative topic and under the constraints of the creative paradigm, the process involves generating planning documents, storyboard documents, and final cuts in sequence to obtain a video file containing the target content.

[0006] In one embodiment, the step of determining at least one creative topic reference material and the event context of the material based on the set of materials to be processed includes: Based on creative needs information, the inspiration search agent retrieves corresponding creative materials from a wide range of media data sources. The creative materials are semantically parsed to extract key elements from each material and construct a structured event graph that describes the relationships between materials and the connections between events within a single material. Based on preset clustering conditions, the creative materials are clustered to determine at least one topic material cluster. The material event context of each topic material cluster is determined from the structured event graph.

[0007] In one embodiment, the step of semantically matching the corresponding material to be processed with the preset material paradigms in the paradigm library for each creative theme, and filtering out high-quality theme materials whose matching degree meets preset conditions, includes: By conducting in-depth research on intelligent agents, feature extraction is performed on reference materials for creative topics to obtain features to be matched. Semantic matching is performed between the features to be matched and the pre-defined paradigm features to determine the semantic similarity of the features to be matched. Creative topic reference materials that meet the preset conditions in terms of semantic similarity will be used as high-quality theme materials.

[0008] In one embodiment, the steps of generating creative topics by calling a pre-trained creative generation model based on high-quality topic materials and their corresponding event context include: Through the topic creative intelligence agent, high-quality topic materials and the material event context corresponding to their respective topic material clusters are obtained to construct creative generation input information; The creative generation input information is fed into a pre-trained creative generation model to generate at least one set of candidate creative topics; the creative generation model is obtained by fine-tuning a general large language model. The candidate creative topics are feature-encoded and quantitatively scored based on creative paradigm constraints. Select at least one target creative topic from the candidate creative topics that meets the preset criteria; the target creative topic includes the topic title, core hook description, value dimension tags, rhythm blueprint and content overview.

[0009] In one embodiment, based on a creative topic and under the constraints of a creative paradigm, the steps of generating a planning document, a storyboard document, and a final cut sequentially to obtain a video file containing the target content include: By using a copywriting planning AI, we can obtain target creative topics and matching scenario-based creative paradigms. Based on the target creative topic and scenario-based creative paradigm, combined with the corresponding scenario-based prompt word templates, a large language model fine-tuned by scenario corpus is used to generate structured planning copy through chain prompts. The intelligent storyboard agent analyzes structured planning documents, uses semantic boundary recognition algorithms to segment scenes, and generates standardized storyboard documents. The intelligent agent, based on the case, reads standardized storyboard documents, calls the video timing generation model to generate video clips shot by shot, and outputs a video file containing the target content through clip splicing, transition addition, audio-visual synchronization and subtitle synthesis.

[0010] In one embodiment, the method further includes: Establish a unified task state data structure for data interaction among multiple intelligent agents; Based on a unified task state data structure, the input and output data between each intelligent agent are standardized and encapsulated. After each agent completes its execution, the corresponding output results are checked for structural integrity and semantic consistency. If the verification result does not meet the preset conditions, the corresponding agent is triggered to regenerate the output result, or roll back to the previous agent to re-execute the corresponding steps; If the verification result meets the preset conditions, the output result of the current agent is sent to the next agent to continue to execute the corresponding video generation step.

[0011] Secondly, to achieve the above objectives, this application further provides a creative video generation apparatus, the apparatus comprising: The video generation module, in response to user creative needs, executes the following steps through multiple intelligent agents: Obtain users' creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed; Based on the set of materials to be processed, identify at least one creative topic reference material and the event context of the material. For each creative topic reference material, semantic matching is performed between the creative topic reference material and the pre-set paradigm features in the paradigm library, and high-quality topic materials that meet the preset conditions are selected from each creative topic reference material. Based on high-quality thematic materials and their corresponding event context, a pre-trained creative generation model is invoked to generate creative topics. Based on the creative topic and under the constraints of the creative paradigm, the process involves generating planning documents, storyboard documents, and final cuts in sequence to obtain a video file containing the target content.

[0012] Thirdly, to achieve the above objectives, this application further provides a creative video generation device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the creative video generation method described above.

[0013] Fourthly, to achieve the above objectives, this application further provides a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the above-described creative video generation method.

[0014] Fifthly, to achieve the above objectives, this application provides a computer program product comprising a computer program that, when executed by a processor, implements the steps of the above-described creative video generation method.

[0015] One or more technical solutions proposed in this application have at least the following technical effects: This application constructs a multi-agent collaborative creative video generation system, achieving fully automated processing from material collection, event context analysis, creative topic generation, planning document generation, storyboard generation to final output, under a unified task state data structure and creative paradigm constraints. Compared to related technologies where agents only transmit coarse-grained text results and lack state inheritance and anomaly recovery mechanisms, this application achieves standardized transmission of generation states through a unified task state data structure. This allows subsequent generation stages to inherit thematic information, emotional characteristics, and rhythmic features from previous stages, effectively reducing issues such as theme deviation, emotional discontinuity, and shot rhythm conflicts. Simultaneously, through paradigm feature semantic matching and quantitative scoring mechanisms, dynamic adaptation of creative topics and generated content under different content types is achieved, improving the stability, consistency, and content quality of creative video generation results, further enhancing the automation level and fairness of creative video generation. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the creative video generation method in the embodiments of this application.

[0019] Figure 2 This is a schematic diagram of the structure of a creative video generation device.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application embodiment is as follows: First, based on user needs, general media material retrieval and semantic analysis are performed to construct a structured event graph and perform topic clustering to form a collection of creative materials with event context; then, a paradigm library is introduced to perform semantic matching and quality screening of the materials to obtain high-quality theme materials, and candidate creative topics are generated by a pre-trained creative generation model in combination with the event context of the materials, and then the target creative topics are selected through quantitative scoring and paradigm constraints; on this basis, structured planning documents, standardized storyboard documents and final video production are generated step by step through intelligent agents such as copywriting planning, intelligent storyboarding and script-based production; at the same time, a unified task state data structure is established to standardize the encapsulation and state inheritance of data among multiple intelligent agents, and a structural integrity and semantic consistency verification and anomaly rollback mechanism are introduced to ensure the continuity, stability and content consistency of the multi-agent collaborative generation process.

[0024] Specifically, this application provides a method for generating creative videos, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the creative video generation method of this application.

[0025] In this embodiment, the creative video generation method includes: In response to the user's creative needs, the following steps S10 to S50 are executed by multiple intelligent agents: Step S10: Obtain the user's creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed.

[0026] Step S20: Based on the set of materials to be processed, determine at least one source material for the creative topic and the context of the material events.

[0027] Step S30: For each type of creative topic reference material, perform semantic matching between the creative topic reference material and the preset paradigm features in the paradigm library, and select high-quality topic materials that meet the preset conditions from each creative topic reference material.

[0028] Step S40: Based on high-quality thematic materials and their corresponding event context, call the pre-trained creative generation model to generate creative topics.

[0029] Step S50: Based on the creative topic and under the constraints of the creative paradigm, the planning document, storyboard document, and final cut are generated sequentially to obtain a video file containing the target content.

[0030] The core idea of ​​this embodiment is to collect relevant materials based on the user's creative needs, summarize the event timelines of each material, and determine high-quality materials that meet the target requirements by combining them with a preset theoretical paradigm. Based on these materials, creative ideas are generated and refined. Finally, by combining text-based video and image-based video technologies, a complete creative video that meets the user's needs is generated. For example, if a user wants to generate a viral video related to the Double Ninth Festival, after understanding the needs, this embodiment first collects materials based on keywords such as "Double Ninth Festival" and determines high-quality materials according to the theoretical paradigm required for viral videos. Thus, by combining the event timelines and characteristics of the materials, a creative topic with high potential for high viewership can be generated. Then, based on this topic and the current theoretical paradigm of viral videos, a creative video with high potential for high viewership can be generated.

[0031] In addition, the steps in this embodiment are executed serially by multiple intelligent agents, which may include an inspiration search intelligent agent, an in-depth research intelligent agent, a topic creation intelligent agent, a copywriting planning intelligent agent, an intelligent storyboard intelligent agent, and a script-based animation intelligent agent.

[0032] Specifically, the system first responds to the user's input of creative requirements and initiates a material collection process based on these requirements. These requirements can include the video theme, content style, and dissemination goals. For example, if a user wants to generate a short video on the theme of "Double Ninth Festival family affection," they can use an intelligent inspiration search agent to search the entire internet for relevant and legally valid materials related to semantic keywords such as "Double Ninth Festival," "family affection," "reunion," and "respect for the elderly," completing the initial information aggregation and contextualization for content creation. Subsequently, the system processes the acquired creative materials in a unified format, performs data cleaning and semantic analysis, forming a set of materials to be processed for subsequent topic refinement and event analysis. This approach avoids the problem of limited material sources in traditional solutions, improving the coverage and richness of the materials.

[0033] Based on the set of materials to be processed, the relationships between materials and the event logic within the materials are analyzed to determine the reference materials for creative topics and the corresponding event context. For each type of reference material for creative topics, semantic matching is performed with pre-set paradigm features in the paradigm library to filter out high-quality thematic materials. The paradigm library can pre-store a large number of creative paradigm features related to viral content, such as viral paradigms that conform to the clock theory, HKRR value system, and other viral logic. Through in-depth research on the intelligent agent, features are extracted from the reference materials for creative topics to obtain features to be matched, and semantic similarity is calculated between these features and pre-set paradigm features. If the corresponding semantic similarity meets the preset conditions, the corresponding reference material for creative topics is identified as high-quality thematic material. For example, for the material about "children who have not returned home for many years returning to their hometown to accompany their elderly parents on the Double Ninth Festival," its emotional resonance, expression of festival value, and emotional progression structure may all have a high degree of matching with viral paradigms, and therefore it can be selected as high-quality thematic material. This method avoids the problem of relying entirely on human experience to select materials in traditional solutions, improving the efficiency and accuracy of identifying high-potential content.

[0034] The system generates creative topics based on high-quality thematic materials and their corresponding event contexts, using a pre-trained creative generation model. This model is trained by fine-tuning a general-purpose language model for specific scenarios, resulting in enhanced creative generation capabilities and content style adaptability. After obtaining candidate creative topics, the system further encodes their features and performs quantitative scoring based on creative paradigm constraints, selecting target creative topics that meet preset criteria. For example, the system can generate topics with emotional contrast and impact, such as "An elderly person says they don't want their child to come home, but secretly prepares a meal," and further output corresponding topic titles, core hook descriptions, value dimension tags, rhythm blueprints, and content summaries. This approach ensures that generated topics are not only creative but also better aligned with the target platform's dissemination patterns and user emotional preferences. Finally, based on the target creative topic and under creative paradigm constraints, the system sequentially generates planning documents, storyboards, and final cuts, outputting a video file containing the target content.

[0035] In one feasible implementation, step S20 includes steps A10 to A40: Step A10: Based on the creative needs information, obtain corresponding creative materials from general media data sources through the inspiration search agent.

[0036] Step A20 involves semantic parsing of the creative materials, extracting key elements from each material, and constructing a structured event graph that describes the relationships between materials and the connections between events within a single material.

[0037] Step A30: Based on preset clustering conditions, cluster the creative materials to determine at least one topic material cluster.

[0038] Step A40: Determine the material event context of each topic material cluster from the structured event graph.

[0039] Step S30 includes steps B10 to B30: Step B10 involves conducting in-depth research on the intelligent agent to extract features from the reference materials for the creation topic and obtain the features to be matched.

[0040] Step B20: Perform semantic matching between the feature to be matched and the pre-defined paradigm features to determine the semantic similarity of the feature to be matched.

[0041] Step B30: Select creative topic reference materials whose semantic similarity meets the preset conditions as high-quality theme materials.

[0042] Step S40 includes steps C10 to C40: Step C10: Through the topic creative intelligence agent, obtain high-quality topic materials and the material event context corresponding to the topic material clusters they belong to, and construct creative generation input information.

[0043] Step C20: Input the creative generation input information into the pre-trained creative generation model to generate at least one set of candidate creative topics; the creative generation model is obtained by fine-tuning the training of a general large language model.

[0044] Step C30: Encode the candidate creative topics by features and quantify the candidate creative topics based on creative paradigm constraints.

[0045] Step C40: Select at least one target creative topic from the candidate creative topics that meets the preset criteria; the target creative topic includes the topic title, core hook description, value dimension tags, rhythm blueprint and content overview.

[0046] Step S50 includes steps D10 to D40: Step D10: Use the copywriting planning AI to obtain the target creative topic and the matching scenario-based creative paradigm.

[0047] Step D20: Based on the target creative topic and scenario-based creative paradigm, and combined with the corresponding scenario-based prompt word templates, a large language model fine-tuned by scenario corpus is used to generate structured planning copy through chain prompts.

[0048] Step D30: The intelligent storyboard agent analyzes the structured planning document, uses a semantic boundary recognition algorithm to segment the scene, and generates a standardized storyboard document.

[0049] Step D40: The intelligent agent based on the case reads the standardized storyboard document, calls the video timing generation model to generate video clips shot by shot, and outputs a video file containing the target content through clip splicing, transition addition, audio-visual synchronization and subtitle synthesis.

[0050] In addition, to ensure the consistency of input and output among the agents, the method also includes steps E10 to E50: Step E10: Establish a unified task state data structure for data interaction among multiple agents.

[0051] Step E20: Based on a unified task state data structure, the input and output data between the agents are standardized and encapsulated.

[0052] Step E30: After each agent has completed its execution, perform structural integrity verification and semantic consistency verification on the corresponding output results.

[0053] In step E40, if the verification result does not meet the preset conditions, the corresponding agent is triggered to regenerate the output result, or the process is rolled back to the previous agent to re-execute the corresponding step.

[0054] In step E50, if the verification result meets the preset conditions, the output result of the current agent is sent to the next agent to continue executing the corresponding video generation step.

[0055] Specifically, after creative requirements enter the system, the task orchestration unit first performs a structured analysis of the requirements and maps the results to corresponding creative tasks. Then, based on the creative task, an inspiration search agent is activated to retrieve, aggregate, and preprocess publicly available materials from a broad media data source. The retrieved text, images, videos, audio, and their associated metadata are uniformly converted into a computable and transferable set of materials to be processed, thus providing basic input for subsequent event analysis and creative generation. This approach allows material collection to move beyond a single source, forming a broader and more comprehensive input pathway, improving the completeness of information in the early stages of creation.

[0056] After obtaining the set of materials to be processed, key elements in each creative material are extracted through semantic parsing, entity recognition, and event extraction, and a structured event graph that can represent the relationship between materials and the event connection within a single material is established.

[0057] Based on preset clustering conditions, the system clusters creative materials to form at least one topic-based material cluster, and extracts the event context of each cluster from the corresponding structured event graph. Thus, the system can not only identify the surface theme of the materials, but also trace the inherent logical relationships between the occurrence, development, and outcome of events. This allows material selection to move beyond manual browsing and subjective judgment, and instead rely on automatic analysis based on calculable event chains.

[0058] For example, the knowledge graph and event context are constructed by generating semantic vectors based on the BERT-base-chinese pre-trained model, extracting core keywords and event triples (subject-behavior-object) through TF-IDF and TextRank algorithms, using dependency parsing to sort out the event logic, and outputting a structured event graph containing all materials.

[0059] The K-Means++ clustering algorithm was used to cluster content based on semantic similarity, publication time, region, and audience tags. A cosine similarity of ≥0.85 was set as the clustering threshold. Duplicate topics were merged, and at least one topic-specific content cluster was output. Then, based on the structured event graph, the event context of each topic-specific content cluster was determined.

[0060] After identifying reference materials for the creative topics, the intelligent agent performs feature extraction on each reference material to obtain features to be matched. These features are then semantically matched with pre-set paradigm features in a paradigm library to obtain corresponding semantic similarity. The paradigm library pre-stores multiple types of paradigm features related to viral content, including rhythm features corresponding to clock theory, value features corresponding to the HKRR value system, and structured content features in different scenarios. High-quality thematic materials are selected based on whether the semantic similarity meets preset conditions. In this way, candidate materials that better conform to the target dissemination pattern and have higher content value density can be automatically identified from a large amount of material, avoiding the problems of inconsistent material selection standards and reliance on experience judgment in traditional solutions.

[0061] In one example, the agent conducts an in-depth study of the theoretical requirements of the HKRR value system to determine the relevant features that need to be extracted from the reference materials for the creation topic, which can reflect happiness (H), knowledge (K), resonance (R), and rhythm (Rh). These features are then matched with the pre-set paradigm features related to the HKRR value system in the paradigm library using cosine similarity. Materials with a matching degree ≥0.7 are judged as high-value materials. The materials are then ranked by weighted order of knowledge scarcity, emotional resonance, and dissemination potential.

[0062] After obtaining high-quality thematic materials and their corresponding event context, the topic creative agent combines the two into creative generation input information and inputs it into a pre-trained creative generation model to generate at least one set of candidate creative topics.

[0063] The creative generation model can be trained based on a general-purpose large language model through scenario-based fine-tuning, thereby possessing stronger content style adaptability and creative organization capabilities. After generating candidate creative topics, the system will also encode the features of the candidate creative topics and perform quantitative scoring under the constraints of the creative paradigm. For example, it will score in real time from dimensions such as hook attractiveness, value density, potential for virality, and scenario adaptability, thereby filtering out target creative topics that meet the preset conditions. Correspondingly, the target creative topic can include a topic title, core hook description, value dimension tags, rhythm blueprint, and content overview, making the creative results not only expressive but also executable and implementable.

[0064] In one implementation, the candidate creative topics output by the creative generation model do not directly enter the subsequent generation stage, but are first screened by a quantitative scoring module. This scoring process can comprehensively consider multiple dimensions such as emotional resonance, knowledge scarcity, repetition rate, rhythmic appeal, scene adaptability, and dissemination potential, making the generated creative topics more in line with the dissemination preferences of different platforms. For example, for creative topics geared towards short video platforms, the system can prioritize content with strong hooks, fast pace, and high emotional impact; for knowledge-based or long-form video content, it can prioritize content with a complete structure, high information density, and clear rhythmic levels. In this way, the relevance of the creative topics and the dissemination effect of the final video content can be significantly improved.

[0065] In steps D10 to D40, based on the target creative topic and its matching scenario-based creative paradigm, the planning document, storyboard document, and final cut are generated sequentially. The copywriting planning agent first acquires the target creative topic and its corresponding scenario-based creative paradigm, and then, combined with scenario-based prompt word templates, uses chained prompts to guide a large language model fine-tuned from scenario corpora to generate a structured planning document. The structured planning document can further include content positioning, overall framework, paragraph arrangement, emotional rhythm, visual style, and script, giving the planning result clear creative guidance. Because the copywriting generation process is constrained by the scenario-based creative paradigm, it effectively reduces problems such as off-topic content, style drift, and loose expression.

[0066] For example, LoRA weights can be trained specifically for different scenarios such as galas, cultural tourism, sports events, and education through LoRA fine-tuning. During the generation of specific planning documents, the corresponding LoRA is invoked based on the input content, improving scenario adaptability. During the generation process, built-in scenario-based prompt templates and chained prompts (CoT) guide the model to generate the document step-by-step, embedding corresponding constraint rules from clock theory and popular theories such as HKRR to prevent AI illusions. The planning document is decomposed into hierarchical levels based on Markdown parsing algorithms, automatically verifying completeness, and triggering model completion for missing modules, conforming to industry standards.

[0067] Subsequently, the intelligent storyboard agent reads the structured planning document and uses a semantic boundary recognition algorithm to segment the document into scenes. Based on the visual shot library, it automatically matches types such as long shot / medium shot / close-up / aerial shot, and generates standardized storyboard documents under the constraints of the creative paradigm (for example, allocating shot duration under the constraints of the creative paradigm corresponding to the clock theory).

[0068] By transforming planning documents into standardized storyboard documents, abstract creative expressions can be converted into actionable shot-level tasks, thereby providing clear input for video timing generation models and improving the stability and controllability of final product generation.

[0069] The intelligent agent reads standardized storyboard documents and uses a video timing generation model to generate video clips shot by shot. These clips are then spliced, transitions added, audio-visual synchronization achieved, and subtitles synthesized, ultimately outputting a video file containing the target content. Because each shot clip originates from the structured constraints of the preceding storyboard documents, the final product maintains overall thematic consistency, harmonious pacing, and coherent visual expression, while also effectively inheriting the value dimensions and emotional blueprint established by the initial creative topic. Through this approach, the system achieves an automated closed-loop generation process from creative topic to final product, appropriately meeting user needs in terms of output content, possessing high playback potential, and reducing the costs of multiple rounds of manual modification and cross-tool switching.

[0070] Furthermore, to ensure data interaction consistency among multiple agents, this embodiment establishes a unified task state data structure in steps E10 to E50, and standardizes and encapsulates the input and output data between agents based on this unified task state data structure. After each agent completes its execution, the system performs structural integrity and semantic consistency checks on its output results. If the check results do not meet preset conditions, the corresponding agent is triggered to regenerate the output results, or the process rolls back to the previous agent to re-execute the corresponding steps. If the check results meet preset conditions, the current agent's output results are sent to the next agent to continue executing subsequent video generation steps. Through this mechanism, the state inheritance and output stability of each agent can be maintained during serial collaboration, avoiding interruptions in the creation process due to missing intermediate results, inconsistent formats, or semantic offsets.

[0071] A unified task state data structure can include at least the following fields: task identifier, execution status, completion progress, current stage input, current stage output, feature vector, next agent identifier, and timestamp. Each agent, upon receiving input, reads the output of the preceding agent from the unified task state data structure and fills its own generated results back into the state structure for subsequent agents to process. In this way, the system can achieve continuous transfer of task state among multiple agents, allowing subsequent generation stages to inherit thematic information, emotional features, and rhythmic features of previous generation stages, thereby effectively reducing problems such as thematic deviation, emotional discontinuity, and conflicting camera rhythms.

[0072] Furthermore, this embodiment also provides a paradigm iteration closed-loop method for evaluating the quality of generated content and, based on the evaluation data and actual dissemination data, reverse-optimizing the viral paradigm library to achieve closed-loop optimization throughout the entire process, specifically including: PPO based on reinforcement learning achieves paradigm self-evolution: The reward value is calculated based on the data dissemination: Reward = f(completion rate, likes, shares, favorites). Constructing the paradigm vector gradient update: ΔPatternVector = α × (Reward - baseline) × LogPolicy The paradigm library code and scoring weights are updated synchronously, and a new paradigm version is automatically built every week to achieve continuous evolution.

[0073] Content Evaluation: Based on the scoring rules of quantitative scoring standard units, the automatically generated content is analyzed in depth from multiple dimensions to generate an overall score, content rating, detailed sub-analysis report, and prediction of the content's market dissemination performance.

[0074] Data collection: This is used to collect real market performance data such as actual playback data, completion rate, interaction data, and cross-platform dissemination effects after the content is launched.

[0075] Paradigm Iteration and Optimization: Based on content evaluation reports and actual market performance data, we reverse-engineer the optimization direction of the top-level design paradigm, continuously improve, enhance and consolidate the creative paradigms and scoring standards in the hit paradigm library, forming a positive closed loop of "paradigm guiding creation - creation producing content - content data reverse-engineering - paradigm optimization and upgrading".

[0076] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the creative video generation method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0077] This application also provides a creative video generation device, which includes: The video generation module, in response to user creative needs, executes the following steps through multiple intelligent agents: Obtain user creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed; Based on the set of materials to be processed, at least one creative topic reference material and the event context of the material are determined; For each creative topic reference material, the creative topic reference material is semantically matched with the preset paradigm features in the paradigm library, and high-quality topic materials that meet the preset conditions are selected from each creative topic reference material. Based on the high-quality thematic materials and their corresponding event context, a pre-trained creative generation model is invoked to generate creative topics. Based on the creative topic and under the constraints of the creative paradigm, the process involves generating planning documents, storyboard documents, and final cuts in sequence to obtain a video file containing the target content.

[0078] The creative video generation apparatus provided in this application, employing the creative video generation method in the above embodiments, can solve the technical problem of unstable creative video generation quality in related technologies. Compared with related technologies, the beneficial effects of the creative video generation apparatus provided in this application are the same as those of the creative video generation method provided in the above embodiments, and other technical features in the creative video generation apparatus are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0079] This application provides a creative video generation device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the creative video generation method in the above embodiments.

[0080] The following is for reference. Figure 2 The diagram illustrates a structural schematic of a creative video generation device suitable for implementing embodiments of this application. The creative video generation device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), tablets, PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 2 The creative video generation device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0081] like Figure 2 As shown, the creative video generation device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 (ROM) or a program loaded from a storage device 1003 into a random access memory 1004 (RAM). The random access memory 1004 also stores various programs and data required for the operation of the creative video generation device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 (I / O interface) is also connected to the bus 1005. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the creative video generation device to communicate wirelessly or wiredly with other devices to exchange data. Although a creative video generation device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0082] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0083] The creative video generation device provided in this application, employing the creative video generation method in the above embodiments, can solve the technical problem of unstable creative video generation quality in related technologies. Compared with related technologies, the beneficial effects of the creative video generation device provided in this application are the same as those of the creative video generation method provided in the above embodiments, and other technical features in this creative video generation device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0084] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0085] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0086] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the creative video generation method in the above embodiments.

[0087] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0088] The aforementioned computer-readable storage medium may be included in the creative video generation device; or it may exist independently and not assembled into the creative video generation device.

[0089] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0090] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0091] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0092] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described creative video generation method, thereby solving the technical problem of unstable creative video generation quality in related technologies. Compared with related technologies, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the creative video generation method provided in the above embodiments, and will not be repeated here.

[0093] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the creative video generation method described above.

[0094] The computer program product provided in this application can solve the technical problem of unstable quality in the generation of creative videos in related technologies. Compared with related technologies, the beneficial effects of the computer program product provided in this application are the same as those of the creative video generation method provided in the above embodiments, and will not be repeated here.

[0095] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for generating creative videos, characterized in that, The method includes: In response to the user's creative needs, the following steps are executed through multiple intelligent agents: Obtain user creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed; Based on the set of materials to be processed, at least one creative topic reference material and the event context of the material are determined. For each creative topic reference material, the creative topic reference material is semantically matched with the preset paradigm features in the paradigm library, and high-quality topic materials that meet the preset conditions are selected from each creative topic reference material. Based on the high-quality thematic materials and their corresponding event context, a pre-trained creative generation model is invoked to generate creative topics. Based on the creative topic and under the constraints of the creative paradigm, the process involves generating planning documents, storyboard documents, and final cuts in sequence to obtain a video file containing the target content.

2. The creative video generation method as described in claim 1, characterized in that, The step of determining at least one creative topic reference material and the event context of the material based on the set of materials to be processed includes: Based on the aforementioned creative needs information, the inspiration search agent retrieves corresponding creative materials from a broad media data source. The creative materials are semantically parsed to extract key elements from each material and construct a structured event graph that describes the relationships between materials and the connections between events within a single material. Based on preset clustering conditions, the creative materials are clustered to determine at least one topic material cluster. The material event context of each topic material cluster is determined from the structured event graph.

3. The creative video generation method as described in claim 1, characterized in that, The step of semantically matching the corresponding materials to be processed with the preset material paradigms in the paradigm library for each creative theme, and filtering out high-quality theme materials that meet the preset matching conditions, includes: By conducting in-depth research on intelligent agents, feature extraction is performed on the reference materials for the creative topics to obtain the features to be matched. The semantic similarity of the features to be matched is determined by performing semantic matching between the features to be matched and the pre-defined paradigm features. Creative topic reference materials whose semantic similarity meets the preset conditions will be used as high-quality theme materials.

4. The creative video generation method as described in claim 2, characterized in that, The steps of generating creative topics based on the high-quality thematic materials, combined with their corresponding event context, and by calling a pre-trained creative generation model include: By using the topic creative intelligence agent, the event context of the high-quality topic materials and their corresponding topic material clusters is obtained, and creative generation input information is constructed. The creative idea generation input information is fed into a pre-trained creative idea generation model to generate at least one set of candidate creative topics; the creative idea generation model is obtained by fine-tuning a general large language model. The candidate creative topics are feature-encoded and quantitatively scored based on creative paradigm constraints. Select at least one target creative topic from the candidate creative topics that meets the preset criteria for scoring; the target creative topic includes a topic title, core hook description, value dimension tags, rhythm blueprint, and content overview.

5. The creative video generation method as described in claim 1, characterized in that, The steps involved in generating a video file containing the target content, based on a creative topic and constrained by a creative paradigm, include: generating a planning document, generating a storyboard, and generating the final cut in sequence. The copywriting planning AI agent is used to obtain the target creative topics and their matching scenario-based creative paradigms. Based on the target creative topic and the scenario-based creative paradigm, and combined with the corresponding scenario-based prompt word templates, a large language model fine-tuned by scenario corpus is used to generate structured planning copy through chain prompts. The intelligent storyboard agent parses the structured planning document, uses a semantic boundary recognition algorithm to segment scenes, and generates standardized storyboard documents. The intelligent agent that generates videos based on specific scenarios reads the standardized storyboard document, calls the video timing generation model to generate video clips shot by shot, and outputs a video file containing the target content through clip splicing, transition addition, audio-visual synchronization and subtitle synthesis.

6. The creative video generation method as described in claim 1, characterized in that, The method further includes: Establish a unified task state data structure for data interaction among multiple intelligent agents; Based on the unified task state data structure, the input and output data between each intelligent agent are standardized and encapsulated. After each agent completes its execution, the corresponding output results are checked for structural integrity and semantic consistency. If the verification result does not meet the preset conditions, the corresponding agent is triggered to regenerate the output result, or roll back to the previous agent to re-execute the corresponding steps; If the verification result meets the preset conditions, the output result of the current agent is sent to the next agent to continue to execute the corresponding video generation step.

7. A creative video generation device, characterized in that, The device includes: The video generation module, in response to user creative needs, executes the following steps through multiple intelligent agents: Obtain user creative needs information, collect materials based on the creative needs information, and obtain a set of materials to be processed; Based on the set of materials to be processed, at least one creative topic reference material and the event context of the material are determined; For each creative topic reference material, the creative topic reference material is semantically matched with the preset paradigm features in the paradigm library, and high-quality topic materials that meet the preset conditions are selected from each creative topic reference material. Based on the high-quality thematic materials and their corresponding event context, a pre-trained creative generation model is invoked to generate creative topics. Based on the creative topic and under the constraints of the creative paradigm, the process involves generating planning documents, storyboard documents, and final cuts in sequence to obtain a video file containing the target content.

8. A creative video generation device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the creative video generation method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the creative video generation method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the creative video generation method as described in any one of claims 1 to 6.

Images