Artificial intelligence content production real-time risk control method and apparatus

Abstract

The present application provides an artificial intelligence content production real-time risk control method and apparatus. The method comprises: step S1, receiving input data needing to be used for synthesizing a digital avatar video; step S2, determining the security and compliance of the input data according to a preset risk control requirement, and if the input data is secure and compliant, releasing a risk control lock applied to a digital avatar model; step S3, synthesizing the digital avatar video on the basis of the input data, and by means of the digital avatar model, performing compliance approval on the synthesized video located in a cache; and step S4, performing disk storage on a final video that has passed compliance approval. In the present application, real-time risk control management can be performed on video content generated by digital avatars, thereby preventing the malicious use of the digital avatars.

Description

A real-time risk control method and device for artificial intelligence content production Technical Field

[0001] The present application belongs to the field of video processing technology, and in particular relates to a real-time risk control method and device for artificial intelligence content production. Background Art

[0002] With the widespread development of generative AI content generation and digital human technology, more and more people are using digital human technology to clone their own images, voices, and brains to create digital avatars. For the first time, digital avatars free humans from the constraints of the physical world. Generative AI content production technology and digital avatars can be used to create videos or live broadcasts that are indistinguishable from those recorded in a physical studio, without the need for real-life interaction. For example, videos featuring live broadcasters can be quickly produced without filming or editing. As the technology matures, the communication effects of digital humans are indistinguishable from those of real people, and viewers will be largely unable to distinguish between real people and digital avatars.

[0003] Digital human technology has significantly advanced the development of livestreaming and internet IP video, but it also poses significant security risks. The existence of digital avatars allows for the creation of videos using their digital counterparts to broadcast content without the knowledge or involvement of the real person. This can cause significant harm. If a digital avatar is used maliciously, it could generate videos that do not reflect the real person's true intentions, causing losses to the real person and, in more serious cases, harm to society.

[0004] Currently, most tools and platforms for the production of generative AI videos using digital humans and digital avatars lack risk management. A few companies utilize keyword filtering to filter scripts before video synthesis, alerting users to potential content risks. Keyword filtering plays a very limited role in the production of digital avatar videos, as keywords can be easily bypassed. In many fields, the content itself is highly specialized, making it difficult to determine its legality through simple text recognition.

[0005] More importantly, in current industry applications, the user of a digital avatar and the real person corresponding to the avatar are likely not the same person. For example, the chairman of a listed company or an expert in the securities and fund industry might have their digital avatars used by employees in the marketing and new media departments, respectively. In such cases, it is crucial to determine whether the content generated by the avatar truly reflects the avatar's intentions. For example, while opinions on hot topics are inherently correct, the user of the avatar and the avatar themselves may disagree, resulting in vastly different video results.

[0006] Summary of the Invention

[0007] In order to solve at least one of the above technical problems, the present application provides a real-time risk control method and device for artificial intelligence content production to perform compliance checks on digital human videos.

[0008] The first aspect of this application is a real-time risk control method for artificial intelligence content production, which mainly includes:

[0009] Step S1: receiving input data required for synthesizing a digital avatar video;

[0010] Step S2: Determine the security compliance of the input data according to the preset risk control requirements. If the input data is safe and compliant, release the risk control lock applied to the digital avatar model.

[0011] Step S3: synthesizing a digital twin video based on the input data, and having the digital twin model perform compliance review on the synthesized video in the cache;

[0012] Step S4: The final video that has passed the compliance review is stored on disk.

[0013] Preferably, step S1 further includes performing a data quality check and a data validity check on the input data.

[0014] Preferably, step S3 further comprises: checking the clarity, fidelity and content of the synthesized video in the cache according to pre-screened video standards.

[0015] Preferably, before step S1, the process further includes pre-approval of the input data by an external risk control system.

[0016] Preferably, after step S4, the method further comprises:

[0017] Step S5: Obtaining the sensitivity of the synthesized video in the cache for compliance review, and simultaneously obtaining the matching degree between the input data and the digital twin. The sensitivity refers to the similarity between the video content and the sensitive content in the compliance review database, and the matching degree refers to the similarity between the input data and the digital twin's keywords.

[0018] Step S6: Determine the audit value Ex based on the compliance similarity and matching degree: Ex=sen-sen*ma;

[0019] Among them, sen is sensitivity and ma is matching degree;

[0020] Step S7: When the audit value Ex exceeds the set value, the video is post-approved by an external risk control system.

[0021] In a second aspect of the present application, a real-time risk control device for artificial intelligence content production mainly includes:

[0022] An input data receiving module, configured to receive input data required for synthesizing a digital avatar video;

[0023] a risk control lock unlocking module, configured to determine the security compliance of the input data according to preset risk control requirements, and to release the risk control lock applied to the digital avatar model if the input data is secure and compliant;

[0024] A video synthesis and approval module, configured to synthesize a digital twin video based on the input data, and have the digital twin model perform compliance approval on the synthesized video in the cache;

[0025] The video output module is used to store the final video that has passed compliance approval on disk.

[0026] Preferably, the input data receiving module further includes a quality inspection unit for performing a data quality inspection and a data legitimacy inspection on the input data.

[0027] Preferably, the video synthesis and approval module further comprises a video standard checking unit for checking the clarity, fidelity and content of the synthesized video in the cache according to pre-approved video standards.

[0028] Preferably, the real-time risk control device for artificial intelligence content production is connected to an external risk control system, and the input data is pre-approved by the external risk control system.

[0029] Preferably, the artificial intelligence content production real-time risk control device further includes:

[0030] A sensitivity and matching acquisition module, configured to acquire the sensitivity of the synthesized video in the cache for compliance review and approval, and to acquire the matching degree between the input data and the digital twin. The sensitivity refers to the similarity between the video content and the sensitive content in the compliance review and approval database, and the matching degree refers to the similarity between the input data and the keywords of the digital twin.

[0031] An audit value calculation module is used to determine an audit value Ex based on the compliance similarity and matching degree: Ex=sen-sen*ma;

[0032] Among them, sen is sensitivity and ma is matching degree;

[0033] The post-approval access module is used to perform post-approval on the video through an external risk control system when the review value Ex exceeds a set value.

[0034] The third aspect of the present application is a computer system comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned real-time risk control method for artificial intelligence content production.

[0035] The fourth aspect of the present application is a readable storage medium, which stores a computer program. When the computer program is executed by a processor, it is used to implement the above-mentioned real-time risk control method for artificial intelligence content production.

[0036] This application can perform real-time risk control management on the video content generated by the digital avatar to prevent the digital avatar from being used maliciously. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] FIG1 is a flow chart of an embodiment of the real-time risk control method for artificial intelligence content production of the present application.

[0038] FIG2 is a schematic diagram of the structure of a computer device suitable for implementing a terminal or server according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the implementation of this application clearer, the technical solutions in the implementation of this application will be described in more detail below in conjunction with the drawings in the implementation of this application. In the drawings, the same or similar numbers throughout represent the same or similar elements or elements with the same or similar functions. The described implementation is a part of the implementation of this application, not all of the implementations. The implementation described below with reference to the drawings is exemplary and is intended to be used to explain this application, and should not be understood as a limitation on this application. Based on the implementation in this application, all other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. The implementation of this application is described in detail below in conjunction with the drawings.

[0040] According to the first aspect of the present application, a real-time risk control method for artificial intelligence content production is provided, as shown in FIG1 , which mainly includes:

[0041] Step S1: receiving input data required for synthesizing a digital avatar video.

[0042] In this step, the input data can be either text or audio. For text, the digital human model can generate speech from it, and for audio, it can also convert it into text. When synthesizing a digital avatar video, the audio input drives the avatar to synthesize lip shapes, expressions, and movements that match the audio, resulting in a video that looks like a real person.

[0043] In some optional implementations, step S1 further includes performing a data quality check and a data validity check on the input data.

[0044] In this embodiment, the quality check part will confirm whether it meets the standards and specifications required by the digital avatar generation content system. For example, the format, resolution, sound quality, etc. of the audio will be checked. If the audio data does not meet the standards, the system will refuse to receive it. Then check whether the content and source of the data are legal, and check whether there are potential risk factors such as malicious injection. If the input data fails the inspection, the system may refuse to continue processing and take appropriate security measures. While performing validity checks on the data input, it prevents malicious injection at the model layer. The system can check the legitimacy of the input data and prevent the injection of malicious data and potential risks.

[0045] Step S2: Determine the security compliance of the input data according to the preset risk control requirements. If the input data is safe and compliant, release the risk control lock applied to the digital avatar model.

[0046] Once the input data passes quality and legality checks, the system will enter the unlocking phase. Risk control locks are a security measure used to restrict the synthesis of digital twin models. If the risk control lock is not released, the digital twin model will remain locked, and video synthesis will not be possible.

[0047] In step S2, the input data is judged for security compliance according to preset risk control requirements, including whether it contains illegal pornography, gambling, or drug-related information, or whether it is false information. If the risk control judgment is passed, the risk control lock is released, and the system will prepare to start synthesizing the digital avatar video.

[0048] It should be noted that the risk control requirements here usually refer to the exclusion of certain words from appearing in the text. This means using keyword recognition technology to identify and filter certain types of words that appear in the text. For example, after building a sensitive word database, automatic review is implemented to determine the security compliance of the input data. This generally includes the following steps:

[0049] (1) Text segmentation: segment the text to be reviewed into words or phrases;

[0050] (2) Keyword matching: Match the words or phrases appearing in the text with the sensitive words in the sensitive word database. If the match is successful, the text is marked as "sensitive information".

[0051] Keyword matching technology mainly refers to calculating similarity, which includes the following steps:

[0052] (21) Determine the keywords of the two sentences to be compared and convert them into feature vectors;

[0053] (22) Select keywords from one of the sentences and traverse the unselected keywords in the other sentence to pair them until all keywords in one of the sentences are paired;

[0054] (23) For each pairing of two sentences, calculate the cosine value of each keyword in the pairing, and calculate the average value as the similarity of the two sentences in this pairing;

[0055] (24) For all two sentences that are paired, the maximum similarity is selected as the similarity of the two sentences.

[0056] For example, sentence 1 consists of three keywords, and the corresponding feature vectors are a1, a2, and a3; sentence 2 consists of three keywords, and the corresponding feature vectors are b1, b2, and b3.

[0057] In the first pairing, the calculated similarity 1 is: v1 = (cos(a1,b1) + cos(a2,b2) + cos(a3,b3)) / 3; in the second pairing, the calculated similarity 1 is: v2 = (cos(a1,b2) + cos(a2,b1) + cos(a3,b3)) / 3; in the third pairing, the calculated similarity 3 is: v3 = (cos(a1,b2) + cos(a2,b3) + cos(a3,b1)) / 3. .... According to the above combination method, a total of 6 similarities can be calculated, and the maximum value is selected as the similarity between the two sentences.

[0058] Afterwards, when the similarity exceeds the audit setting value, the sentence is considered to contain sensitive information, the risk control lock is kept locked, and the situation is fed back to the upper-level user for modification.

[0059] Step S3: synthesize the digital avatar video according to the input data, and the digital avatar model performs compliance review on the synthesized video in the cache.

[0060] The key technology in this application lies in real-time risk control during the digital avatar video synthesis process, integrating the built-in risk control management unit and the digital avatar model into a complete and organic whole. This structure is like a real person in the real world, who will simultaneously judge risks during the filming and performance process to make the next move.

[0061] It should be noted that risk control management using external keyword matching technologies, such as pre-qualification, is referred to as pre-qualification. This pre-qualification risk control method has the advantages of being simple in principle, easy to understand, and low in implementation cost. Since generative AI generally requires a long time and consumes a certain amount of computing power to produce final results, this pre-qualification management method, when applied to risk control in generative AI content production, not only allows for rapid feedback and modification to improve efficiency, but also reduces the waste of electricity and computing power caused by invalid content, thus saving social costs. However, this risk control method has significant drawbacks. First, risk control in the pre-qualification input system is easily circumvented in practice. Whether due to system vulnerabilities, testing environments, or internal management exemptions, these pre-qualification risk controls can be ineffective, such as using commas to isolate sensitive words. Second, during the generative AI content production process, much of the content is generated independently by the AI ​​itself, and the final result may differ significantly from expectations. Risk control based solely on pre-qualified input data poses significant risks.

[0062] To this end, this application further processes the text in the video file formed after the digital avatar model processes the input data to ensure the risk control management of the output video. In step S3, once the digital avatar model is unlocked, it starts to accept input data and generates lip shapes, expressions and movements based on the audio content to synthesize the digital avatar video. After the synthesis is completed, the video content is also subject to compliance review before being stored on the disk. The compliance review is consistent with the principle of judging the security compliance of the input data in step S2. It also uses keyword technology to check false information, illegal content or other security risks. If there is a problem, the system may terminate the synthesis and take corresponding security measures. The entire process is performed in the cache and will not be permanently stored on the disk, avoiding the risk of content being exported or transferred.

[0063] In some optional implementations, step S3 further includes: checking the clarity, fidelity, and content of the synthesized video in the cache according to pre-screened video standards.

[0064] Step S4: The final video that has passed the compliance review is stored on disk.

[0065] Once the synthesized digital avatar video passes approval, the system outputs the generated video content for permanent disk storage. The output video can then be used in appropriate applications, such as posting on social media or live streaming. If it fails approval, the system rejects the output and may trigger an alarm or notify relevant personnel for action.

[0066] The above steps constitute the workflow of the runtime security risk control system, which aims to ensure that the synthesis of digital avatar videos is safe, legal, and consistent with genuine intent. This comprehensive risk control management approach can help prevent bad behavior and potential risks.

[0067] The risk control management model involved in steps S2 and S3 of this application is integrated with the digital twin model, ensuring the consistency and effectiveness of risk control in different application scenarios. Regardless of how the digital twin is upgraded or migrated, the built-in risk control management unit, as part of the trusted digital twin component, will be upgraded and migrated at the same time. In the case of large-scale use of digital twins, in order to speed up the hardware transformation into a dedicated integrated chip, the built-in risk control will be synchronously solidified as part of the dedicated integrated chip. The risk control lock restricts the operation of the digital twin model and can only be unlocked after passing the risk control judgment, thereby improving the security of the system. The security constructed in this way cannot be bypassed at the level of authority management or system operation. Whether in the R&D or testing phase, whether it is an administrator or an ordinary user, even a system attack hacker will be restricted.

[0068] In some optional implementations, before step S1, the process further includes pre-approving the input data through an external risk control system.

[0069] In order to better achieve a balance between system performance, security, and risk control recall rate, it is possible to add pre-emptive risk control to this application. Pre-emptive risk control can be connected to the external risk control system of the enterprise or government, which can more flexibly customize risk control rules according to enterprise needs and configure the latest risk control factors to improve risk control recall rate. Pre-emptive risk control mainly includes two aspects:

[0070] (1) Content keyword review module: Perform keyword review on the text and audio data generated by the digital avatar to determine whether these pre-synthesized materials contain illegal content such as pornography, gambling, and drugs.

[0071] (2) Semantic analysis module: Analyze the semantics of the content to determine whether the script content that will constitute the core of the video contains sensitive information, such as political sensitivity, religious and ethnic sensitivity, etc.

[0072] In addition to adding pre-approval risk control, you can also configure post-approval as needed. Post-approval is used to undertake complex external approval and audit mechanisms and can match advanced authority management and control systems. Post-approval includes the following aspects:

[0073] (1) The content of the production is notified to the production authorized approver or the real person corresponding to the digital avatar defined by the system through SMS and WeChat mini-program notification information,

[0074] (2) The authorized reviewer or the real person corresponding to the digital avatar will check the final result audio, video and other content;

[0075] (3) If the expression is confirmed to be genuine, the risk control system will allow the video to be exported after approval. If the resulting video is not approved, the system can give an approval opinion and reject the generated content. The risk control system will destroy the produced content. The producer can restart the production process based on the approval opinion.

[0076] It is understandable that the post-approval system is to conduct content approval for the final result, which has the risk of high accuracy and no content effect deviation. However, the latter approval system is characterized by low efficiency and high cost, because once the content production is completed, time and resources have been consumed. In addition, this approval method also has the problem of chaotic management of the test environment and digital avatar permissions, the result content is leaked before the result review, or the result fails the review but is still recorded, etc., which leads to the outflow of erroneous result content due to poor management. In this case, the risk is most likely to occur when the producer and the digital avatar are not the same person. To this end, the present application provides an on-demand configuration of the post-approval mechanism. For example, in some optional implementations, after step S4, it further includes:

[0077] Step S5: Obtaining the sensitivity of the synthesized video in the cache for compliance review, and simultaneously obtaining the matching degree between the input data and the digital twin. The sensitivity refers to the similarity between the video content and the sensitive content in the compliance review database, and the matching degree refers to the similarity between the input data and the digital twin's keywords.

[0078] Step S6: Determine the audit value Ex based on the compliance similarity and matching degree: Ex=sen-sen*ma;

[0079] Among them, sen is sensitivity and ma is matching degree;

[0080] Step S7: When the audit value Ex exceeds the set value, the video is post-approved by an external risk control system.

[0081] The sensitivity calculation in step S5 is similar to the similarity calculation described in steps S2 and S3. The difference is that the audit setting values ​​in steps S2 and S3 are higher than the setting value in step S7. The audit setting value in step S2 is based on the sensitivity level and is used to determine whether the text or video should be terminated. When the calculated sensitivity is less than the audit setting value, the text or video content can be circulated, but at this time it is necessary to further judge its size with the setting value in step S7 to determine whether it needs to perform post-review. In this criterion, the matching degree between the input data and the digital avatar is introduced. This is usually associated with the digital avatar model. Any digital avatar model is created by a specific IP, and its association is determined by specified keywords. Whether a to-be-processed input data matches these keywords is the key point for determining the matching between the input data and the digital avatar model. Similarly, the matching degree ma can be calculated based on keyword matching technology, and then the audit value Ex is calculated through step S6. According to the formula of step S6, it can be seen that if an input data has a higher sensitivity and a lower matching degree with the digital avatar model, it is more likely to exceed the setting value of step S7, and it is even more necessary to add a post-approval system to ensure that the input data is an expression of the true meaning of a real person.

[0082] This application addresses the security and compliance issues currently encountered in the application of digital twin technology. By cleverly integrating pre-, post-, and runtime risk control methods, it ensures the authenticity and compliance of digital twin content.

[0083] The second aspect of the present application provides an artificial intelligence content production real-time risk control device corresponding to the above method, mainly comprising:

[0084] An input data receiving module, configured to receive input data required for synthesizing a digital avatar video;

[0085] a risk control lock unlocking module, configured to determine the security compliance of the input data according to preset risk control requirements, and to release the risk control lock applied to the digital avatar model if the input data is secure and compliant;

[0086] A video synthesis and approval module, configured to synthesize a digital twin video based on the input data, and have the digital twin model perform compliance approval on the synthesized video in the cache;

[0087] The video output module is used to store the final video that has passed compliance approval on disk.

[0088] In some optional implementations, the input data receiving module further includes a quality inspection unit configured to perform a data quality inspection and a data validity inspection on the input data.

[0089] In some optional implementations, the video synthesis and approval module further includes a video standard checking unit for checking the clarity, fidelity, and content of the synthesized video in the cache according to pre-approved video standards.

[0090] In some optional implementations, the real-time risk control device for artificial intelligence content production is connected to an external risk control system, and the input data is pre-approved by the external risk control system.

[0091] In some optional implementations, the artificial intelligence content production real-time risk control device further includes:

[0092] A sensitivity and matching acquisition module, configured to acquire the sensitivity of the synthesized video in the cache for compliance review and approval, and to acquire the matching degree between the input data and the digital twin. The sensitivity refers to the similarity between the video content and the sensitive content in the compliance review and approval database, and the matching degree refers to the similarity between the input data and the keywords of the digital twin.

[0093] An audit value calculation module is used to determine an audit value Ex based on the compliance similarity and matching degree: Ex=sen-sen*ma;

[0094] Among them, sen is sensitivity and ma is matching degree;

[0095] The post-approval access module is used to perform post-approval on the video through an external risk control system when the review value Ex exceeds a set value.

[0096] According to the third aspect of the present application, a computer system includes a processor, a memory, and a computer program stored in the memory and runnable on the processor. The processor executes the computer program to implement the above-mentioned real-time risk control method for artificial intelligence content production.

[0097] According to the fourth aspect of the present application, a readable storage medium stores a computer program, which, when executed by a processor, is used to implement the above-mentioned real-time risk control method for artificial intelligence content production.

[0098] 2, which shows a schematic diagram of a computer device 800 suitable for implementing the embodiments of the present application. The computer device shown in FIG2 is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0099] As shown in FIG2 , the computer device 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 708 into a random access memory (RAM) 803. Various programs and data required for the operation of the device 800 are also stored in the RAM 803. The CPU 801, ROM 802, and RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0100] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, and the like; an output section 807 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 808 including a hard disk; and a communication section 809 including a network interface card such as a LAN card or a modem. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 810 as needed, so that computer programs read therefrom can be installed into the storage section 808 as needed.

[0101] In particular, according to the embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 809, and / or installed from the removable medium 811. When the computer program is executed by the central processing unit (CPU) 801, the above-mentioned functions defined in the method of the present application are executed. It should be noted that the computer storage medium of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a 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, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such a propagated data signal may take a variety of forms, including, but not limited to, electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wire, optical cable, RF, etc., or any suitable combination thereof.

[0102] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code includes one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or operation, or can be implemented by a combination of dedicated hardware and computer instructions.

[0103] The modules or units described in the embodiments of this application may be implemented in software or hardware. The modules or units described may also be provided in a processor, and the names of these modules or units do not, in certain circumstances, limit the modules or units themselves.

[0104] The computer-readable storage medium provided in the fourth aspect of this application may be included in the apparatus described in the above embodiments, or may exist independently and not be incorporated into the apparatus. The computer-readable storage medium carries one or more programs, and when the one or more programs are executed by the apparatus, the data is processed according to the above method.

[0105] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A real-time risk control method for artificial intelligence content production, characterized in that: include: Step S1, receiving input data required for synthesizing a digital avatar video; Step S2: Determine the security compliance of the input data according to the preset risk control requirements. If the input data is safe and compliant, release the risk control lock applied to the digital avatar model. Step S3, synthesizing a digital twin video according to the input data, and having the digital twin model perform compliance review on the synthesized video in the cache; Step S4: The final video that has passed the compliance review is stored on disk.

2. The real-time risk control method for artificial intelligence content production according to claim 1, characterized in that: Step S1 further includes performing a data quality check and a data legality check on the input data.

3. The real-time risk control method for artificial intelligence content production according to claim 1, characterized in that: Step S3 further includes: checking the clarity, fidelity and content of the synthesized video in the cache according to pre-screened video standards.

4. The real-time risk control method for artificial intelligence content production according to claim 1, characterized in that: Before step S1, the process further includes pre-approving the input data through an external risk control system.

5. The real-time risk control method for artificial intelligence content production according to claim 1, characterized in that: After step S4, the method further comprises: Step S5, obtaining the sensitivity of compliance review of the synthesized video in the cache, and obtaining the matching degree between the input data and the digital twin, wherein the sensitivity refers to the similarity between the video content and the sensitive content in the compliance review database, and the matching degree refers to the similarity between the input data and the keywords of the digital twin; Step S6: Determine the audit value Ex based on the compliance similarity and matching degree: Ex = sen-sen*ma; Among them, sen is sensitivity, ma is matching degree; Step S7: When the review value Ex exceeds the set value, the video is post-approved by an external risk control system.

6. A real-time risk control device for artificial intelligence content production, characterized in that: include: An input data receiving module, used for receiving input data required for synthesizing a digital avatar video; The risk control lock unlocking module is used to determine the input data according to the preset risk control requirements. Security compliance: if the input data is security compliant, the risk control lock imposed on the digital twin model is released; A video synthesis and approval module, used to synthesize the digital twin video according to the input data, and the digital twin model performs compliance approval on the synthesized video in the cache; The video output module is used to store the final video that has passed the compliance approval on disk.

7. The real-time risk control device for artificial intelligence content production according to claim 6, characterized in that: The input data receiving module also includes a quality inspection unit, which is used to perform a data quality inspection and a data legality inspection on the input data.

8. The real-time risk control device for artificial intelligence content production according to claim 6, characterized in that: The video synthesis and approval module also includes a video standard checking unit, which is used to check the clarity, fidelity and content of the synthesized video in the cache according to the pre-examined video standards.

9. The real-time risk control device for artificial intelligence content production according to claim 6, characterized in that: The real-time risk control device for artificial intelligence content production is connected to an external risk control system, and the input data is pre-approved by the external risk control system.

10. The real-time risk control device for artificial intelligence content production according to claim 6, characterized in that: The artificial intelligence content production real-time risk control device also includes: A sensitivity and matching degree acquisition module, used to acquire the sensitivity of the synthesized video in the cache for compliance review, and at the same time, acquire the matching degree between the input data and the digital twin, wherein the sensitivity refers to the similarity between the video content and the sensitive content in the compliance review database, and the matching degree refers to the similarity between the input data and the keywords of the digital twin; An audit value calculation module is used to determine the audit value Ex based on the compliance similarity and matching degree: Ex = sen-sen*ma; Among them, sen is sensitivity, ma is matching degree; The post-approval access module is used to perform post-approval on the video through an external risk control system when the review value Ex exceeds a set value.

Images