Method and device for generating video, method and device for multi-dimensional right confirmation of video, and medium

Abstract

The application discloses a generative video management and control method, a video multi-dimensional right confirmation method, a device and a medium. The generative video management and control method comprises the following steps: embedding a multi-dimensional right confirmation element in a video generation process, obtaining a target generative video, encapsulating the target generative video as a gateway-controlled rendering data stream package, and the rendering data stream package has a corresponding video API; when a calling party requests the video API, triggering a blockchain smart contract to perform information verification of the calling party through a trusted data space; only when the verification is passed, the encrypted rendering data stream package is issued based on an optical visual countermeasure mechanism of a rendering layer for preventing physical recording; and if the verification fails, the API is automatically cut off. The method can realize productization of the video API, and only when the identity information of the calling party is verified, the encrypted rendering data stream package carrying the optical countermeasure mechanism is issued, so that the naked eye can watch the video without any problem, but the video obtained after shooting is blurred and distorted, loses the value of dissemination, and realizes that people can watch without damage and shooting will be scrapped.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a generative video control method, a video multidimensional rights confirmation method, device, and medium. Background Technology

[0002] With the rapid development of artificial intelligence (AIGC) technology, the threshold for digital video generation has been greatly reduced, driving an explosive growth in the circulation of video assets and reshaping the video creation and dissemination ecosystem. However, it has also given rise to a series of prominent pain points in terms of technology and business.

[0003] Regarding copyright ownership and content protection, generated videos are easily edited, have their frames extracted, or used directly as training material for AI models. Traditional protection methods commonly used in the industry, such as single watermarks and file hash verification, are easily rendered ineffective by routine modifications such as video cropping, compression, and AI redrawing, making it extremely difficult to determine the ownership of video content. At the same time, current anti-recording solutions on the market also have significant shortcomings. Most platforms rely solely on software-level protection mechanisms, using operating system security interfaces to prevent screenshots and screen recording. These methods have significant vulnerabilities; they are not only technically easy to bypass but can also be easily circumvented. Physical methods such as users directly photographing the screen with their phones or using hardware capture cards to intercept video signals can easily bypass software restrictions, leading to the theft and illegal distribution of video content. Summary of the Invention

[0004] In view of this, the present disclosure provides a generative video management and control method, a multi-dimensional video rights confirmation method, an apparatus, and a medium, which can solve the problems of insufficient anti-counterfeiting capabilities for video copyright confirmation and easy bypassing of anti-theft recording protection by hardware and physical means in the field of artificial intelligence video assets.

[0005] In a first aspect, embodiments of this disclosure provide a generative video control method, including: Based on the original input information for video creation, multi-dimensional rights confirmation elements are embedded in the video generation process to obtain the target generative video; The target generative video is encapsulated into a rendering data stream package controlled by the gateway, and the rendering data stream package has a corresponding video API; When a caller requests the video API, the blockchain smart contract is triggered through the trusted data space to verify the caller's information. If the caller's information is verified, the encrypted rendering data stream packet is sent based on the optical visual anti-physical recording mechanism of the rendering layer; If the caller's information verification fails, the video API will automatically cut off the connection.

[0006] Secondly, this disclosure also provides a method for multi-dimensional video rights confirmation, including: In response to the video rights confirmation request signal, the feature matching rate corresponding to the multi-dimensional rights confirmation features is extracted from the generative video to be analyzed. The generative video generation method to be analyzed includes: based on the original input information of video creation, embedding multi-dimensional rights confirmation elements in the video generation process to obtain the target generative video; The authority matching information for the generative video is determined based on the matching rates of all the aforementioned features. If the rights confirmation and matching information meets the preset conditions, the generative video to be analyzed is determined to be a video that illegally misappropriates protected original video elements.

[0007] Thirdly, this disclosure also provides a computer device, which adopts the following technical solution: The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor; the instructions are executed by the at least one processor to enable the at least one processor to perform either the generative video control method or the video multidimensional rights confirmation method described above.

[0008] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium storing computer instructions; the computer instructions are used to cause a computer to execute any of the generative video control methods or the video multi-dimensional rights confirmation methods described above.

[0009] Fifthly, embodiments of this disclosure also provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the methods described above.

[0010] The generative video control method provided in this disclosure embeds multi-dimensional ownership confirmation elements into the video generation process based on the original input information of video creation to obtain a target generative video. The target generative video is then encapsulated into a rendering data stream package controlled by a gateway, with the rendering data stream package having a corresponding video API. When a caller requests the video API, a blockchain smart contract is triggered through a trusted data space to verify the caller's information. If the caller's information verification passes, an encrypted rendering data stream package is sent based on an optical visual countermeasure mechanism to prevent physical recording at the rendering layer; otherwise, the video API is automatically disconnected. This method embeds multi-dimensional ownership confirmation elements during the video generation stage to complete source ownership marking, uniformly manages the video rendering data stream package through a gateway, and automatically verifies the caller's identity using a trusted data space and a blockchain smart contract. If the verification is valid, an encrypted data stream is sent based on an optical visual countermeasure mechanism to resist screen recording; if the verification fails, the transmission link is directly disconnected.

[0011] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the generative video control method provided in an embodiment of the present disclosure.

[0014] Figure 2 This is a flowchart illustrating a method for acquiring a target generative video according to an embodiment of the present disclosure.

[0015] Figure 3 This is a flowchart illustrating the video multidimensional rights confirmation method provided in this embodiment of the disclosure.

[0016] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. Detailed Implementation

[0017] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0018] It should be understood that the following specific examples illustrate the implementation of this disclosure, and those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific implementation methods, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0019] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0020] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The drawings only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0021] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0022] Reference Figure 1 This application discloses a generative video control method, including: S100, based on the original input information for video creation, embeds multi-dimensional rights confirmation elements during the video generation process to obtain a target-generated video.

[0023] S200 encapsulates the target generative video into a gateway-controlled render data stream package, which has a corresponding video API.

[0024] S300: When a caller requests a video API, the blockchain smart contract is triggered through the trusted data space to verify the caller's information.

[0025] S400: If the caller's information is verified, an encrypted rendering data stream packet is sent based on the optical visual anti-physical recording mechanism of the rendering layer.

[0026] S500: If the caller's information verification fails, the video API will automatically cut off the connection.

[0027] The generative video control method disclosed in this application is a novel control method based on trusted data space encapsulation. It refuses to provide physical streaming media file downloads, encapsulating protected AIGC videos and their elements into a rendering data stream package with a corresponding video API controlled by a gateway. When a caller requests the video API, a blockchain smart contract must be triggered through the trusted data space to verify the caller's information. Only after the caller's identity information is verified is an encrypted rendering data stream package carrying optical countermeasures mechanisms issued. Relying on the persistence of human vision and high-frequency alternating modulation in the temporal domain, it ensures that the video is flicker-free, color-free, and without image quality loss when viewed with the naked eye. However, the video obtained after being captured by a physical camera is blurred, distorted, and loses its broadcast value, achieving lossless viewing for humans but immediately rendered unusable after capture. Simultaneously, the countermeasures are embedded in the underlying rendering stage, rather than through post-processing filters, achieving simultaneous encrypted transmission and visual anti-counterfeiting, effectively preventing the leakage of the original unprotected video stream.

[0028] Reference Figure 2 The specific methods for obtaining S100 target generative video include: S110, obtain the original input information for video creation.

[0029] The original input information for video creation includes at least the source video stream, the original creation prompt (i.e., the original creation prompt), copyright holder information, and source video metadata. The source video metadata preferably includes the total number of frames and the total duration of the video.

[0030] For the video streams used in the creation, the local editing software scene can directly open the corresponding project file, locate the original material based on the material reference path provided in the project, retrieve the source file through the material panel, read the uncompressed original bitstream cache in the software cache directory, and check the complete material clips against the timeline track.

[0031] The AI ​​video generation platform allows creators to log in to the creator backend and retrieve the bound reference materials using the corresponding generation task ID. It can also call the platform's open API to retrieve the original binary video stream uploaded by the user, and retrieve material copies from the browser's local upload cache or local storage folder. For real-time materials from cameras and live streaming, the original recording files can be extracted from the shooting device's memory card, or the original streaming slices can be retrieved from the live streaming service provider's backend. If there is no query entry on the front end, a packet capture tool can be used to capture the material upload data packet and export the complete original video stream as a fallback evidence.

[0032] For obtaining the original prompt, the AI ​​video generation tool first retrieves the complete and unmodified original positive and negative prompt words for a single task from the creator's creation history in the creator's backend. It can also call the platform's task query interface to extract the original text of the prompt words at the time of submission from the complete request body. At the same time, it can also parse the local cache of browsers and other sources to restore the input text.

[0033] The prompt text for manually edited scenes includes storyboards, voice-over scripts, etc. The original text can be extracted from the text layer and narration panel within the editing project. At the same time, the matching script documents and the original text content that has not been edited or modified in the platform's cloud draft box can be found. The original text submitted for the first time is retained throughout the process, and the content modified in the later stage is removed.

[0034] In this embodiment, the copyright holder information is preferably a distributed digital identity (DID) string, which can be obtained by querying and retrieving the copyright holder's exclusive information through public data interfaces that are open to the outside world or trusted data space.

[0035] S120 generates a set of secure frame locations using source video metadata and system clock as seeds.

[0036] This step specifically includes: 1) Obtaining the hash value of the source video metadata. Preferably, the total number of video frames and the total video duration are concatenated into a complete text string. The SHA-256 hash algorithm is used to perform calculations on the concatenated string to obtain a fixed-length metadata hash value.

[0037] 2) The hash value and system clock are used together as the initialization seed of the pseudo-random number generator to dynamically obtain several confidential frame positions that are discretely distributed on the video timeline, forming a unique set of confidential frame positions.

[0038] Specifically, the system clock timestamp, accurate to the millisecond level, is collected synchronously at the moment of execution. The binary bit stream of the metadata hash value is concatenated with the binary bit stream of the system clock timestamp. The concatenated complete binary sequence is used as the initialization seed of the pseudo-random number generator to complete the generator initialization.

[0039] The required number of confidential frames is preset, and the pseudo-random number generator that has been initialized is continuously called to generate random floating-point numbers in the range of 0 to 1. The random floating-point numbers are mapped to the corresponding frame numbers in combination with the total number of video frames. The mapping formula is that the frame number is equal to the product of the random floating-point number rounded down from 1 and the total number of frames. Each time a set of frame numbers is generated, the value is checked to see if it falls within the legal range of 1 to the total number of frames and whether it is a duplicate of the already generated frame numbers. Duplicate or out-of-range values ​​are discarded and a new random number is generated until a specified number of non-repeating valid frame numbers are collected. All compliant frame numbers are integrated together to form a unique set of confidential frame locations.

[0040] In this embodiment, when the total number of frames or the total duration of the target video changes, the metadata hash value will change synchronously. Even if the video parameters remain unchanged, the millisecond difference in the system clock at different execution times will also change the initialization seed, ultimately generating a completely different set of confidential frame positions, ensuring that the confidential frame position is bound to the target video, the single generation result is unique, and it is difficult for a third party to forge or replicate it.

[0041] Without a dynamic seed, the pseudo-random number generator outputs fixed frame positions. Attackers can predict all verification frame coordinates in advance and selectively tamper with or delete interpolated verification frames, thus compromising the video asset encryption and verification logic. A dual-source dynamic seed—dynamically changing clock and statically unique video hash—ensures that frame positions are distributed irregularly each time, making discrete interpolation points unpredictable and resisting attacks that target specific tampering or removal of verification frames. The random numbers output by the seed are mapped to the total number of frames and the duration time series, determining the distribution logic of the random number value range. This ensures that all calculated secure frame positions fall within the valid video timeline, preventing invalid interpolation coordinates that exceed the beginning or end of the video and guaranteeing the continuity of the subsequent AIGC interpolation time series context.

[0042] S130: Input the source video stream into the AIGC conditional generation model, and generate the first-stage video containing the encrypted original verification frame based on the set of confidential frame positions.

[0043] This step specifically includes: inputting the source video stream into the AIGC conditional generation model, using the set of confidential frame positions as the time-series generation benchmark, and generating original verification frames corresponding to each confidential frame position based on the discrete frame number in the set. That is, generating one-to-one matching original verification frames at the time axis positions corresponding to the source video stream.

[0044] Each original verification frame is encrypted, with hash processing being the preferred method for encryption, resulting in a first-stage video containing multiple sets of temporal verification frames. The original verification frames after encryption are collectively referred to as temporal verification frames. This first-stage video carries both the spatial domain image corresponding to the pixel image of the complete frame and the visual feature hash generated by hash processing.

[0045] Assuming the total number of frames in the source video stream is 9600, and the previously generated set of confidential frame locations is {428,1563,2907,3611,4592,5106,6039,6671,7245,7880,8412,9157}, after the complete source video stream is continuously fed into the AIGC conditional generation model, the model uses this set of frame numbers as the temporal reference to locate twelve corresponding temporal points such as frame 428 and frame 1563 in the source video stream, and generates the original verification frame matching each point one by one.

[0046] After all the original check frames are generated, extract all pixel spatial image data of each original check frame separately, perform SHA-256 hash operation frame by frame to generate corresponding visual feature hash values, and use each frame image after generating hash identifiers as a time-series check frame.

[0047] Then, all the timing verification frames are arranged and integrated according to the original video timing order, and finally the first-stage video containing twelve sets of timing verification frames is produced. This video fully preserves the original spatial domain pixel image of each verification frame, and each frame is bound to a unique and tamper-proof visual feature hash.

[0048] S140, the first-stage video is spatially converted to the frequency domain, the frequency domain implicit dense points generated based on the copyright information are embedded into the underlying frequency information of the image, and the second-stage video is obtained after inverse transformation to restore the pixel image.

[0049] The steps specifically include: converting and scrambling the copyright information to generate frequency domain implicit dense points, i.e., element A; performing spatial-to-frequency domain conversion on the first-stage video to obtain intermediate frequency domain data; embedding the frequency domain implicit dense points into the underlying image frequency information of the intermediate frequency domain data; and performing restoration processing on the embedded intermediate frequency domain data to obtain the second-stage video.

[0050] The method for obtaining frequency domain implicit dense points specifically includes: converting copyright holder information into a binary data stream, scrambling the order of the binary data stream, and generating a set of pseudo-random dense point perturbation sequences, denoted as frequency domain implicit dense points. That is, after obtaining the copyright holder's DID characters, it is first converted into a computer-recognizable binary data stream, and then the order of these data is scrambled using a chaotic system and scrambling algorithm. For example, a chaotic system such as Logistic chaotic mapping is used to build a scrambling algorithm. Random position indices are generated based on the chaotic sequence, and the original order of the binary bit stream is scrambled according to the indices to eliminate the correlation between the original data. Finally, the scrambled bit stream is mapped to the perturbation values ​​of pixels or frame bits according to preset rules to obtain a set of irregular dense point perturbation sequences.

[0051] A two-dimensional discrete cosine transform is performed on each complete pixel spatial domain image of the time-series verification frame in the first stage video frame by frame to complete the conversion of the image spatial domain to the frequency domain. Specifically, all pixel grayscale and color space information are converted to the frequency domain dimension, and low-frequency, mid-frequency and high-frequency components are obtained to form intermediate frequency domain data.

[0052] The generated frequency domain implicit dense points are embedded into the frequency domain coefficients of the preset low-frequency or high-frequency encrypted channel of the image, and the corresponding channel frequency domain coefficients are modified. A two-dimensional discrete cosine transform is performed on the modified frequency domain image to restore the spatial domain pixel image, thus obtaining the second-stage generative video after embedding the frequency domain implicit dense points.

[0053] Furthermore, the low-frequency information region at the bottom layer of the image, which has the strongest stability and is difficult to detect tampering with the naked eye, is selected first from the intermediate data in the frequency domain. The pre-generated implicit dense points in the frequency domain are then embedded into the low-frequency coefficients in an orderly manner with a fixed step size. During the embedding process, the low-frequency coefficient values ​​are only slightly adjusted to avoid destroying the visual effect of the picture and the original visual feature hash. After all the dense points are embedded, the updated set of frequency domain coefficients is subjected to inverse discrete cosine transform to restore the frequency domain data back to the pixel spatial domain image. All processed time-series verification frames are reassembled and spliced ​​according to the original video time sequence to finally obtain the second-stage video with complete copyright implicit embedding.

[0054] Assuming the original copyright holder information is "Copyright belongs to: XX Digital Media Co., Ltd., Authorization No. SQ20260508001, Original Registration No. DK2026011209", the entire text is first uniformly encoded into a UTF-8 binary byte string. A permutation matrix generated based on the previously mentioned secret frame seed is used to scramble the byte order, completing the out-of-order conversion. Then, the obfuscated binary bits are mapped to small frequency domain offsets in groups of 4 bits, generating a set of continuous frequency domain implicit keypoints, i.e., element A. These keypoints are entirely derived from the original copyright information encoding conversion, with no external redundant data access, providing original data support for subsequent seamless and highly robust embedding.

[0055] After generating the dense points, frequency domain preprocessing is performed on the original video in the first stage to avoid the inherent defects of traditional RGB three-channel DCT transformation. This method abandons the conventional approach of directly modifying the RGB three channels synchronously in blocks using DCT. Based on the characteristics of human vision, the original RGB color space of the video frame is uniformly converted to the BT.601 standard YUV420 color space. Taking advantage of the human eye's sensitivity to changes in the luminance channel and its insensitivity to changes in the chrominance channel, only the luminance Y channel is selected as the sole dense point embedding carrier. The original pixel data of the U and V chrominance channels are completely preserved without any frequency domain transformation or numerical modification. This completely solves the problem of visible distortion caused by the luminance shift of the G channel and the color shift superposition of the R / B channels in the RGB scheme.

[0056] Specifically, the selected Y-channel image is divided into 8×8 non-overlapping uniform blocks, and a two-dimensional DCT transformation is performed on each block to separate the DC coefficients and low-frequency AC coefficients of each block. All block frequency domain data are integrated to form distortion-free video frequency domain intermediate data that is adapted to dense point embedding.

[0057] In the key point embedding stage, CRC cyclic redundancy error correction coding is first performed based on the complete key point sequence of element A. The encoded key point data packet is split in a fixed ratio of 7:3. 70% of the core body offset is used for DC coefficient embedding, and 30% of the check redundancy data is equally divided into three groups.

[0058] During embedding, the DC coefficients of a single block are no longer modified independently. Instead, the original DC coefficients of the four neighboring blocks above, below, left, and right of the current block are extracted in real time. The average brightness of the neighborhood is calculated, and the scaling factor α is adaptively matched with the human eye brightness recognition threshold of 4 as the judgment standard. For flat texture areas such as the sky and solid color walls, a low scaling factor of 0.2 to 0.4 is used to strictly limit the modification range of the DC coefficients and ensure that the difference between the corrected DC coefficient and the neighborhood average is lower than the human eye perception threshold, thus completely eliminating the brightness discontinuity between blocks and the grid block effect. For complex texture areas, a high scaling factor of 0.7 to 1.0 is used to ensure the embedding strength of dense points.

[0059] The scaled offset of the main dense point is superimposed on the low-frequency coefficients of the DC layer of each block. At the same time, three sets of verification redundancy data are sequentially embedded into the (0,1), (1,0), and (1,1) low-frequency AC coefficients of each block adjacent to the DC. The low-frequency AC redundancy compensates for the watermark energy deficiency caused by the limited modification range of the DC in the flat area. The robustness of copyright verification of the dense point can be guaranteed without amplifying the offset. This solves the inherent contradiction of traditional schemes where a very small offset is traceless but ineffective, and a large offset is robust but distorted.

[0060] In this embodiment, only minor adjustments are made to the values ​​of the underlying low-frequency coefficients throughout the process. The overall pixel distribution, color characteristics, and brightness logic of the video frames are not changed, and the original visual feature hash is fully preserved. There are no color blocks, blurs, color shifts, or grid distortions that can be recognized by the naked eye.

[0061] After all frequency domain blocks are embedded, inverse DCT transformation is performed on each of the 8×8 frequency domain blocks to restore the processed complete Y channel pixel image. The Y channel is then merged with the original U and V chroma channels without any modification, and then converted in reverse to a standard RGB pixel image to complete the restoration of a single frame image.

[0062] Specifically, the complete process of color space conversion, Y-channel block DCT, adaptive dense point layer embedding, and inverse DCT image restoration is repeatedly performed on all time-series verification frames corresponding to the 12 confidential locations in the video set. Finally, according to the original frame time sequence of 428, 1563, 2907...9157, all processed verification frames are integrated to output the second-stage video with embedded copyright frequency domain hidden points, no visible distortion, complete preservation of the original visual hash verification features, and strong robust copyright verification capabilities.

[0063] S150, in the second stage of the video execution graphics rendering pipeline, overlays the dynamic space layer mask onto the core target in each frame of the image to obtain the target-generated video.

[0064] The steps specifically include: 1) analyzing the brightness information of each frame in the source video stream, adaptively adjusting the local transparency of the corresponding pattern, and obtaining a dynamic space layer mask; 2) in the second stage of video execution graphics rendering pipeline stage, identifying the core target in each frame image; 3) superimposing the dynamic space layer mask onto the core target to obtain a target-generated video.

[0065] The method for obtaining the dynamic spatial layer mask specifically includes: acquiring each frame of the video image, using a lightweight object detection model or a salient region detection model to analyze and locate the single frame image, and identifying and outputting the region of interest corresponding to the core visual object within the frame.

[0066] Invoke the preset system-level encryption matrix, and generate a geometric masking pattern with spatial geometric pseudo-random motion characteristics and a gray level lower than the critical visual perception threshold of the human eye based on the system-level encryption matrix.

[0067] Calculate the local average brightness and local contrast corresponding to each region of interest. Based on the calculated local average brightness and local contrast, perform adaptive local transparency adjustment on the geometric masking pattern, with the transparency value constrained within a preset range. The geometric masking pattern after adaptive transparency adjustment is the dynamic spatial layer mask.

[0068] Subsequently, the second-stage video, which already has embedded frequency domain implicit dense points and carries visual feature hashes, is sent to the GPU graphics rendering pipeline. During the rasterization stage, core targets such as people, products, and scene subjects in the image are identified frame by frame, and the coordinates and motion trajectory of the corresponding target region of interest, i.e., the core target ROI, are output.

[0069] At the fragment shader processing level, the pre-generated dynamic spatial layer mask is read. Based on the identified core target ROI, the overlay area is locked. The mask layer is then applied to the core target surface with a preset extremely low opacity parameter and moves synchronously with the target. Spatial binding marks are completed on the surface of the video image. That is, the mask coordinates are updated synchronously with the target displacement and scaling changes in each frame to achieve dynamic alignment.

[0070] In this application, the underlying video image, frequency domain hidden points and visual hash information are not destroyed throughout the process. The layer overlay only completes the spatial binding mark on the surface of the image. After processing all the timing verification frames, the final target generative video is output by integrating according to the original video timing.

[0071] Assuming the source video is a product demonstration video, and the core target of the screen is a smart device, first extract the pixel brightness of each frame of the source video frame by frame. The average brightness of the highlight area of ​​the device is 230, and the corresponding local transparency of the dynamic mask is preferably set to 0.75. The average brightness of the shadow area of ​​the device is 60, and the corresponding local transparency of the dynamic mask is set to 0.15. Divide the blocks to generate a dynamic spatial layer mask that adapts to the brightness and darkness, i.e., element D.

[0072] The second-stage video, containing embedded copyright frequency domain implicit key points and visual feature hashes for each verification frame, is fed into the graphics card rendering pipeline. The pipeline's real-time target detection model selects the ROI region of the smart device frame by frame and tracks the device's translation and rotation trajectories. A dynamic spatial layer mask is loaded in the GPU fragment shader, and the mask layer is superimposed only within the device ROI. The shader corrects the mask position in real time to match the device's dynamic changes. The overlay transparency uses the adaptively calculated partition parameters. The overlay operation only applies to the surface layer of the image, and the frequency domain implicit key points and visual feature hashes stored at the bottom layer are completely preserved without modification. The mask dynamic overlay rendering is performed sequentially on twelve confidentiality time-series verification frames from 428, 1563, 2907 to 9157. After all frames are processed, they are spliced ​​and integrated in the order of the original video timeline to finally obtain a target-generated video with dynamic spatial binding marks on the surface and copyright secrets and verification hashes at the bottom layer.

[0073] In this application, the AIGC conditional generation model can be a video frame-level image conditional generation model, such as StableDiffusion, SDXL, Midjourney, Flux, etc., which can receive the original image pixels of the corresponding confidential frame position in the source video as spatial conditions and the frame number time sequence as temporal conditions, and generate original verification frames with the same scene and composition based on the structure of the source image. Alternatively, it can be a dedicated video generation diffusion model that natively supports video stream input and timeline frame index constraints, and can output original verification frames matching the scene frame by frame, strictly according to the temporal points of the confidential frame position set, based on the source video stream as a reference. It should be noted that any existing model that can meet the video generation requirements of this application is acceptable and is within the scope of protection of this application.

[0074] After the target generative video is generated, this application also includes: obtaining the semantic feature hash space of the material, i.e., element C, based on the material video stream and the original creation Prompt; encapsulating the association information of the original video creation input information and the corresponding encrypted original verification frame, i.e., element B, frequency domain implicit key point, i.e., element A, semantic feature hash space, i.e. element C, and dynamic space layer mask, i.e. element D, into on-chain evidence storage metadata, and uploading it to the blockchain. Specifically, the on-chain evidence storage metadata is uploaded to the blockchain, and all features of elements A, B, C, and D are permanently fixed in an immutable distributed ledger through a smart contract.

[0075] The method for obtaining the semantic feature hash space, i.e., element C, specifically includes: extracting the text semantic vector corresponding to the original creation Prompt, i.e., the prompt word string; identifying keyframes in the source video stream and obtaining the visual feature hash of the keyframes; obtaining the semantic feature hash space based on the text semantic vector and the visual feature hash, i.e., performing modal fusion of the obtained text semantic vector with the visual hash sequence of all keyframes, mapping the static image hash, dynamic color motion feature hash, and text semantic features to the same high-dimensional semantic feature hash space, and constructing a standard semantic hash space matrix in matrix format according to frame time sequence and semantic dimension.

[0076] After the matrix is ​​generated, the matrix, along with metadata such as video ownership, creation prompt, and hash generation parameters, is encrypted, packaged, and uploaded to the corresponding blockchain in the trusted data space for storage and solidification.

[0077] The method for obtaining the visual feature hash of keyframes specifically includes: parsing the source video stream frame by frame, selecting keyframe images that carry the main subject, composition information, core color and color motion features of the image, and constructing a macroscopic semantic picture of the corresponding video based on all keyframe images; performing standardized preprocessing on each selected keyframe image, which includes one or more of the following: size scaling, grayscale conversion, noise reduction and smoothing.

[0078] Choose either the perceptual hashing algorithm or the differential hashing algorithm to perform global visual feature extraction and compression encoding on the preprocessed grayscale image of a single keyframe, generating a binary encoding string of fixed bit length; use the binary encoding string as the unique visual feature hash corresponding to the keyframe to complete the acquisition of the visual feature hash of a single keyframe.

[0079] The S200 encapsulates the target generative video into a rendering data stream package with a corresponding video API controlled by the gateway, thus productizing the video API. It refuses to provide physical streaming media file downloads and encapsulates the protected AIGC video and its elements into a rendering data stream package controlled by the gateway, communicating with the outside world through the video API.

[0080] In this step, the encapsulation specifically includes: establishing a one-to-one binding relationship between the rendering data stream package and the dedicated video API; after the target generative video is generated, allocating a globally unique data stream identifier and registering the corresponding video API access entry; the rendering service initializes the data stream package template; and writing the data stream identifier and API identifier into the package header to complete the static association.

[0081] Furthermore, the rendering data stream packet can be divided into a header control field, a rendering media payload field, and a packet tail verification field. It uses Protobuf for binary serialization encapsulation. The gateway can achieve control by parsing only the packet header without decoding the audio and video payload. The packet header carries the data stream number, API mapping identifier, control instructions, tenant authentication information, media parameters, segment sequence number, and running status.

[0082] For S300-S500, when a caller requests a video API, the caller's information must be verified through a blockchain smart contract triggered by the trusted data space. This verification includes checking the caller's account balance and authorization scope. If the verification is successful, the encrypted rendering stream will be sent in real time; if the verification fails, the API will be automatically disconnected.

[0083] Specifically, when the caller's information verification passes, the encrypted rendering data stream packet is sent based on the optical visual anti-physical recording mechanism of the rendering layer. This includes: sending the encrypted rendering data stream packet to the client corresponding to the caller; during the graphics pipeline rendering process on the client, dynamically injecting anti-physical recording features into the underlying graphics rendering pipeline to obtain and present the actual rendering brightness.

[0084] The temporal modulation of the anti-physical recording feature (i.e., adversarial signal) includes: injecting a high-frequency alternating signal into the luminance channel or a specific color channel in the continuously output video frame sequence, setting the original pixel luminance of the t-th frame, dynamically generating a small deviation threshold, and obtaining the actual rendered luminance after injecting the adversarial signal. : , Let be the original pixel brightness of frame t. This is a dynamically generated threshold for minute deviations.

[0085] This method utilizes the persistence of vision in the human eye. The alternating positive and negative ∆L is averaged by the human eye's biological filter at high refresh rates, while discrete sampling physical cameras will capture severe flicker due to shutter misalignment. Relying on the persistence of vision in the human eye and high-frequency alternating modulation in the time domain, the positive and negative brightness shifts are... The human eye will automatically smooth out the image, so the video appears flicker-free, color-free, and without any loss of image quality when viewed with the naked eye. However, physical cameras use discrete shutter sampling, and if the shutter speed and image refresh are misaligned, obvious flicker noise will be captured. The video will be blurry and distorted after being copied, losing its value for dissemination. This achieves lossless viewing for humans, but is immediately rendered unusable after being captured.

[0086] Conventional protection only blocks system screenshot and screen recording software, and cannot prevent physical recording behavior such as mobile phones and cameras directly shooting the screen. This solution dynamically injects optical countermeasure features into the underlying graphics rendering pipeline, without relying on upper-layer software interception, and interferes with external camera shooting from the image imaging level, covering physical copying infringement scenarios that software protection cannot solve, and providing more complete protection.

[0087] Furthermore, to prevent piracy from using frame-by-frame averaging algorithms for post-processing restoration, spatial frequency band interleaving injection can be implemented. Specifically, the image is divided into several interlaced blocks, such as a checkerboard array. Within the same frame t, odd-numbered blocks apply +∆L brightness compensation, and even-numbered blocks apply -∆L brightness compensation. After frame t+1, a polarity reversal operation is performed, effectively preventing piracy from restoring the image using frame-by-frame averaging algorithms. By employing a checkerboard-grid block positive and negative brightness compensation and a spatial interleaving mechanism that reverses the polarity of adjacent frames, even if piracy uses frame-by-frame averaging and noise reduction restoration algorithms, they cannot eliminate the inherent adversarial textures in the image. This significantly increases the cost of reverse restoration, watermark removal, and flicker removal, making the protection more robust and difficult for post-processing image processing methods to remove adversarial features.

[0088] In this embodiment, the global average brightness of the current rendering frame can be analyzed in real time, and the absolute value of ∆L can be dynamically adjusted. For example, the offset can be slightly reduced for bright scenes and moderately increased for dark scenes, always keeping the adversarial signal within the threshold invisible to the human eye, so that there will be no visual problems such as graying in bright scenes or flickering in dark scenes. It can be stably effective for both high-brightness scenes, dark-toned movies and TV shows, and low-brightness text and video, and is suitable for all types of video rendering scenarios.

[0089] In this application, the adversarial features are directly injected into the client's graphics rendering pipeline, without relying on dedicated display devices or external anti-reproduction hardware. Optical interference is achieved solely through software rendering logic, making it compatible with various terminals with graphics rendering, such as PCs, mobile devices, and large screens. It has low deployment costs and a wide range of compatibility. The strobe and checkerboard textures generated by the camera reproduction have unique identifiable features. Later, image algorithms can be used to quickly determine whether the video is a physically recorded version, which can serve as an objective technical basis for copyright infringement determination, facilitating platform review and judicial evidence collection.

[0090] For example, for a standard 60Hz refresh rate display device, the client receives the API data stream for local rendering. Analyzing the current video frame as having a medium grayscale, the adversarial strength ∆L is calculated to be 4%. During the rendering of the first frame, the image is divided into 16x16 checkerboard blocks, with the brightness of all odd-numbered blocks increased by 4%, and the brightness of even-numbered blocks decreased by 4%. During the rendering of the second frame, approximately 16.6ms later, the brightness of odd-numbered blocks decreases by 4%, and the brightness of even-numbered blocks increases by 4%, and this cycle repeats frequently.

[0091] After adopting this method, for legitimate users, due to the critical flicker fusion frequency characteristics of the human eye, the visual nerves of the user will cancel out the brightness deviation between the previous and next frames, i.e., +4% and -4%, when watching, resulting in a smooth and stable perceived picture.

[0092] For illegal recording devices, when a thief uses a mobile phone camera at 30fps to film the screen, the camera's rolling shutter and progressive scan characteristics mean its exposure cycle cannot be aligned with the screen's refresh rate. This physical hardware interference causes the recorded MP4 file to exhibit not only severe overall flickering between light and dark areas, but also large, coarse black-and-white moiré patterns (water ripples) that move with the scan lines. This physical damage occurs the instant the image sensor is exposed to light and cannot be repaired by software in post-production, effectively stripping the recorded video of its dissemination value.

[0093] Reference Figure 3 Secondly, this application discloses a method for multi-dimensional rights confirmation via video, including: S10, in response to the video rights confirmation request signal, extracts the feature matching rates corresponding to the multi-dimensional rights confirmation features from the generative video to be analyzed.

[0094] The generation method of the generative video to be analyzed includes: based on the original input information of video creation, embedding multi-dimensional rights confirmation elements in the video generation process to obtain the target generative video.

[0095] The original input information for video creation associated with each target generative video, the corresponding encrypted original verification frame, frequency domain implicit key points, semantic feature hash space, dynamic spatial layer mask, and other related information are all encapsulated as on-chain evidence metadata and stored on the blockchain.

[0096] It should be noted that the specific method for obtaining the target-generated video in this embodiment is consistent with the method for obtaining the target-generated video disclosed in the first aspect of this application, so it will not be described in detail here.

[0097] In this embodiment, the multidimensional rights confirmation features include at least two of the following: frequency domain implicit key points, time-series verification frames (i.e., encrypted original verification frames), semantic feature hash space, dynamic spatial layer mask, and on-chain evidence storage metadata.

[0098] The feature matching rates corresponding to the multi-dimensional rights confirmation features include at least two of the following: frequency domain implicit dense point matching rate, time sequence verification frame matching rate, semantic feature hash space matching rate, dynamic spatial layer mask matching rate, and on-chain evidence storage metadata matching rate.

[0099] Specifically, the method for obtaining the frequency domain implicit dense point matching rate includes: performing frame unpacking processing on the generative video to be analyzed to obtain continuous video frame images; performing a two-dimensional discrete cosine transform on each frame image that is completely consistent with the dense point embedding stage to transform the image from the spatial domain to the frequency domain.

[0100] Locate the same preset low-frequency or high-frequency encrypted channel and read the frequency domain coefficients inside the channel; according to the system's preset threshold rules, for example, if the coefficient shows positive fine-tuning, it is recorded as binary 1, and if the coefficient shows negative fine-tuning, it is recorded as binary 0, thereby blindly extracting a set of binary streams to be tested.

[0101] Retrieve the exclusive decryption key that the copyright holder has pre-uploaded to the trusted data space unified platform and registered on the blockchain. Use the decryption key to perform reverse chaotic scrambling and restoration processing on the binary stream under test to obtain the restored identity string. Compare the restored identity string with the copyright holder's standard DID string stored on the blockchain using Hamming distance or global string alignment comparison, and count the number of valid decryption bits that are correctly demodulated.

[0102] The preset total number of injected dense point bits in the frequency domain implicit dense point is retrieved during the embedding stage. The percentage corresponding to the ratio of the number of correctly demodulated valid dense point bits to the total number of injected dense point bits is used as the frequency domain implicit dense point matching rate.

[0103] The method for obtaining the timing verification frame matching rate specifically includes: the verification end sending a retrieval request to the blockchain trusted data space to retrieve the security metadata corresponding to the target video asset number; parsing and extracting the confidential time series set corresponding to the implicit verification frames of the original video from the security metadata, which includes k preset total verification frames; and based on each timestamp in the confidential time series set, performing precise fixed-point frame extraction on the video to be analyzed to extract candidate frame images corresponding to each of the k timestamps.

[0104] Then, the compliance of candidate frames is verified in two dimensions. The first dimension uses an optical flow algorithm or a temporal continuity verification model to detect the visual coherence and image feature similarity between each candidate frame and its adjacent context frames, determining the temporal visual coherence compliance of the candidate frames. The second dimension extracts the encrypted hash value of each candidate frame and compares it one by one with the hash templates of the temporal verification frames registered in the blockchain security metadata to determine hash compliance. The number of valid verification frames that simultaneously meet the conditions of successful hash collision comparison and temporal visual coherence is obtained. The percentage corresponding to the ratio of the number of valid verification frames to the preset total number of verification frames k is calculated as the temporal verification frame matching rate of the video to be analyzed.

[0105] The method for obtaining the semantic feature hash space matching rate specifically includes: inputting the generative video to be analyzed into a preset multimodal feature extraction network, performing temporal decoding to split the complete video into a continuous frame sequence, extracting keyframes carrying the main subject, composition, core color, and color motion features of the image according to the keyframe selection rules of the original video, extracting the global main subject action trajectory corresponding to each keyframe, and aggregating the image features of all keyframes to obtain the macroscopic image semantics of the video to be analyzed. The multimodal feature extraction network fuses the generated macroscopic image semantic picture and the temporal main subject action trajectory features to obtain a video fusion visual feature set that integrates static image information and dynamic motion information. Using a differential hashing algorithm of the same origin as the original video, global visual feature extraction and compression encoding are performed on the fusion visual feature set to generate a fixed bit length binary code. The binary codes corresponding to all keyframes are collected and encapsulated in temporal order to finally obtain the perceptual differential hash set corresponding to the entire video to be analyzed.

[0106] Secondly, within the blockchain nodes of the trusted data space, the standard semantic hash space matrix corresponding to the protected original video elements stored on the chain is retrieved. Using the Hamming distance mathematical model or cosine similarity formula, the hash space distance between the perceptual difference hash set of the video to be analyzed and the standard semantic hash space matrix of the original video—two sets of high-dimensional hash sets—is calculated and denoted as DH.

[0107] Finally, the semantic feature hash space matching rate is obtained. The semantic feature hash space matching rate is the value of 1-DH / N{bits}, where N{bits} is the total number of fixed bits in the hash encoding.

[0108] The method for obtaining the dynamic spatial layer mask matching rate specifically includes: performing target object detection on the video to be analyzed, defining the target region of interest corresponding to the core visual subject in each frame of the current image, and accurately locking the target carrying area that needs to be compared with the mask trace.

[0109] Subsequently, the encrypted seed that is uniformly shared at the bottom layer when the dynamic spatial layer mask is generated is retrieved. Based on this seed, the complete state of the standard dynamic mask within the corresponding video time interval is independently and synchronously deduced and reproduced in memory, restoring the spatial geometric deformation law, pseudo-random displacement trajectory and gray level distribution of the original mask, and completely replicating the standard mask layer that has not been tampered with.

[0110] Microscopic pixel grayscale distribution data within the ROI region of each frame of the video to be analyzed are extracted. At the same time, pixel data of the corresponding coordinate region of the standard mask layer synchronously reproduced in memory are retrieved. Bidirectional pixel comparison is completed by spatial cross-correlation coefficient calculation or pixel-by-pixel residual statistics to quantify the degree of overlap between the two types of pixel arrays.

[0111] If the video to be analyzed has not undergone tampering operations such as partial cropping, image mirroring, or partial erasure, the pixels in the core visual subject ROI area will retain subtle spatial fluctuation traces caused by the micro-transparent geometric masking pattern after adaptive transparency adjustment. Tampering will destroy the pixel correspondence and reduce the degree of matching overlap.

[0112] Finally, based on the spatial cross-correlation coefficient or pixel residual statistical results, the percentage of pixel spatial geometric trajectory matching overlap is calculated. This percentage value is the dynamic spatial layer mask matching rate, which fully characterizes the degree of matching between the original dynamic mask layer of the video to be analyzed and the residual masking traces in the current image.

[0113] The method for obtaining the on-chain evidence storage metadata matching rate specifically includes: when the video to be analyzed triggers the rights confirmation process within the system, extracting its integrated asset digital fingerprint header information. Using this header information as an index key, a distributed ledger data status query is initiated to the Trusted Data Space Blockchain Consortium Network. Due to the technical characteristics of blockchain's absolute immutability and global trust, this verification result is presented in a binary logic, i.e., existence or non-existence, alignment or misalignment. If the corresponding smart contract is successfully addressed and the read metadata pointer is unaltered, it is directly confirmed as 100%; if there is no such evidence storage on the chain or the asset fingerprint is completely broken and cannot be indexed, it is confirmed as 0%.

[0114] S20, determine the weighting and matching information of the generative video to be analyzed based on the matching rates of all features.

[0115] Specifically, the weighted summation of all feature matching rates yields the weighted matching information for the generative video.

[0116] It should be noted that the number of feature matching rates can be flexibly selected according to actual needs. When calculating the weighted matching information, it is sufficient to ensure that the sum of the weights of all selected matching rates is 1.

[0117] S30, if the rights matching information meets the preset conditions, the generative video to be analyzed is determined to be a video that illegally misappropriates the elements of the protected original video.

[0118] For example, if a suspected stolen video with a black-and-white filter and reduced frame rate is discovered, the multi-dimensional rights confirmation features extracted using this method correspond to the following feature matching rates: frequency domain implicit dense point matching rate, temporal verification frame matching rate, semantic feature hash space matching rate, dynamic spatial layer mask matching rate, and on-chain evidence metadata matching rate. These five matching rates are 60%, 50%, 95%, 90%, and 100%, respectively, with configured weights of 0.3, 0.2, 0.2, 0.15, and 0.15. Therefore, the rights confirmation matching information S of the generative video to be analyzed is S = 0.6 × 0.3 + 0.5 × 0.2 + 0.95 × 0.2 + 0.9 × 0.15 + 1.0 × 0.15 = 0.755. Assuming a preset threshold of 0.75, since 0.755 > 0.75, it indicates that the generative video to be analyzed has been stolen and belongs to the category of stolen videos. This method can still successfully confirm rights even if the video has been significantly damaged or altered.

[0119] Furthermore, this application can be applied to the dynamic revenue distribution of multi-dimensional rights confirmation in generative videos. Specifically, when a target generative video generates new derivative APIs (child nodes) and generates commercial revenue, the smart contract can capture the five-tuple feature traces of the parent node in the child node; the comprehensive matching score S of the child node to the parent node is used as the revenue sharing calculation factor; when the smart contract deducts revenue each time the child node API generates revenue, it automatically transfers the funds to the original author's wallet with zero delay, based on the actual sharing ratio of the actual sharing calculation model, i.e., the contribution rate × the matching score S.

[0120] Existing video copyright transactions are mostly buyouts or platform subscriptions, lacking standardized mechanisms and making it impossible to finely control the use of individual materials and differentiate rights. Furthermore, AIGC creation often involves the stitching together and secondary creation of materials from multiple sources. Current technology cannot quantify the proportion of the original video in the newly generated video, meaning that the contribution of multiple source materials in AIGC creation is difficult to quantify. By using the method of public disclosure in this application, we can effectively achieve the rights confirmation analysis of AIGC created videos and accurately determine the contribution of the creator.

[0121] A computer device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0122] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the computer device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory, causing the computer device to perform all or part of the steps of the generative video control method or video multi-dimensional rights confirmation method of the foregoing embodiments of this disclosure.

[0123] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0124] like Figure 4 This is a schematic diagram of a computer device provided for an embodiment of the present disclosure. It illustrates a structural schematic diagram suitable for implementing the computer device in the embodiments of the present disclosure. Figure 4 The computer device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0125] like Figure 4 As shown, a computer device may include a processor (such as a central processing unit, graphics processing unit, etc.), which performs various appropriate actions and processes based on programs stored in read-only memory (ROM) or programs loaded from storage devices into random access memory (RAM). The RAM also stores various programs and data required for the operation of the computer device. The processor, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0126] Typically, the following devices can be connected to the I / O interface: input devices, such as sensors or visual information acquisition devices; output devices, such as displays; storage devices, such as magnetic tapes or hard drives; and communication devices. Communication devices allow the computer device to communicate wirelessly or wiredly with other devices (such as edge computing devices) to exchange data. Although Figure 4A computer apparatus with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or included alternatively.

[0127] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory 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 a storage device, or installed from a ROM. When the computer program is executed by a processor, all or part of the steps of the generative video control method or the video multi-dimensional rights confirmation method of embodiments of this disclosure are performed.

[0128] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0129] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When the non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the generative video control method or video multi-dimensional rights confirmation method of the foregoing embodiments of the present disclosure are performed.

[0130] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0131] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0132] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0133] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0134] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0135] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0136] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0137] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0138] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A generative video control method, characterized in that, include: Based on the original input information for video creation, multi-dimensional rights confirmation elements are embedded in the video generation process to obtain the target generative video; The target generative video is encapsulated into a rendering data stream package controlled by the gateway, and the rendering data stream package has a corresponding video API; When a caller requests the video API, the blockchain smart contract is triggered through the trusted data space to verify the caller's information. If the caller's information is verified, the encrypted rendering data stream packet is sent based on the optical visual anti-physical recording mechanism of the rendering layer; If the caller's information verification fails, the video API will automatically cut off the connection.

2. The generative video control method according to claim 1, characterized in that, The process of embedding multi-dimensional rights confirmation elements into the video generation process based on the original input information for video creation to obtain the target generative video includes: Obtain the original input information for video creation, which includes the source video stream, original creation prompts, copyright holder information, and source video metadata; A set of secure frame locations is generated using the source video metadata and system clock as seeds; The source video stream is input into the AIGC conditional generation model, and the first-stage video containing the encrypted original verification frame is generated using the set of confidential frame positions as the time-series generation benchmark. The first stage video is processed by spatial-to-frequency domain conversion, and the frequency domain implicit dense points generated based on the copyright holder information are embedded into the underlying frequency information of the image. After inverse transformation to restore the pixel image, the second stage video is obtained. In the second stage, the video execution graphics rendering pipeline stage, a dynamic space layer mask is overlaid on the core target in each frame to obtain a target-generated video.

3. The generative video control method according to claim 2, characterized in that, The process of generating a set of secure frame locations using the source video metadata and system clock as seeds includes: Obtain the hash value of the metadata of the source video; The hash value and the system clock are used together as the initialization seed for the pseudo-random number generator to dynamically obtain several confidential frame positions that are discretely distributed on the video timeline, forming a unique set of confidential frame positions.

4. The generative video control method according to claim 2, characterized in that, The step of inputting the source video stream into the AIGC conditional generation model, using the set of confidential frame positions as the temporal generation benchmark, to generate a first-stage video containing encrypted original verification frames includes: The source video stream is input into the AIGC conditional generation model, and the original verification frames corresponding to each of the confidential frame positions are generated using the set of confidential frame positions as the time-series generation benchmark. The original verification frame is encrypted to obtain the first stage video.

5. The generative video control method according to claim 2, characterized in that, The step of performing spatial-to-frequency domain conversion on the first-stage video, embedding the frequency domain implicit dense points generated based on the copyright holder information into the underlying frequency information of the image, and then performing an inverse transformation to restore the pixel image to obtain the second-stage video includes: The copyright holder information is converted and scrambled to generate frequency domain implicit dense points; The first stage video is spatially converted to the frequency domain to obtain intermediate frequency domain data. The implicit dense points in the frequency domain are embedded into the underlying frequency information of the image in the intermediate frequency domain data. The intermediate frequency domain data after embedding is then restored to obtain the second-stage video.

6. The generative video control method according to claim 2, characterized in that, In the second stage of the video execution graphics rendering pipeline, a dynamic space layer mask is superimposed on the core target in each frame to obtain a target-generative video, including: Analyze the brightness information of each frame in the video stream and adaptively adjust the local transparency of the corresponding pattern to obtain a dynamic spatial layer mask; In the second stage, the video execution graphics rendering pipeline stage, the core targets in each frame of the image are identified; The dynamic spatial layer mask is overlaid on the core target to obtain a target-generated video.

7. The generative video control method according to claim 2, characterized in that, After the target generative video is generated, the following is also included: The semantic feature hash space of the original generative video asset is obtained based on the source video stream and the original creation prompts; The original input information for video creation, along with the associated information of the corresponding encrypted original verification frame, the frequency domain implicit key point, the semantic feature hash space, and the dynamic spatial layer mask, are encapsulated into on-chain evidence storage metadata and uploaded to the blockchain.

8. The generative video control method according to claim 7, characterized in that, The step of obtaining the semantic feature hash space of the original generative video asset based on the source video stream and the original creation prompts includes: Extract the text semantic vector corresponding to the original creation prompt words; Identify keyframes in the source video stream and obtain the visual feature hashes of the keyframes; A semantic feature hash space is obtained based on the text semantic vector and the visual features.

9. A method for multi-dimensional rights confirmation of video, characterized in that, include: In response to the video rights confirmation request signal, the feature matching rate corresponding to the multi-dimensional rights confirmation features is extracted from the generative video to be analyzed. The generative video generation method to be analyzed includes: based on the original input information of video creation, embedding multi-dimensional rights confirmation elements in the video generation process to obtain the target generative video; The authority matching information for the generative video is determined based on the matching rates of all the aforementioned features. If the rights confirmation and matching information meets the preset conditions, the generative video to be analyzed is determined to be a video that illegally misappropriates protected original video elements.

10. A computer device, characterized in that, The computer device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor; the instructions are executed by the at least one processor to enable the at least one processor to perform the generative video control method according to any one of claims 1-8 or the video multidimensional rights confirmation method according to claim 9.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions; the computer instructions are used to cause the computer to execute the generative video control method according to any one of claims 1-8 or the video multidimensional rights confirmation method according to claim 9.

12. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the generative video control method according to any one of claims 1-8 or the video multidimensional rights confirmation method according to claim 9.

Images