Method and system for automatically identifying and splitting animation of PPT composite video
Through deep learning models and neural network technology, the PPT animation effects are analyzed, and the difficulty of complex animation recognition and splitting is solved, the automation and animation effects of video generation are improved, and the video production process is simplified.
Patent Information
- Application Number
- CN202411249700.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-13
AI Technical Summary
The prior art has difficulty in identifying and splitting when processing complex animation effects in PPT, which affects the smoothness and authenticity of the animation, and has low degree of automation, resulting in inefficient video generation and may be distorted in animation effects.
The pre-trained deep learning large language model analyzes the slide file, extracts the visual and timing characteristics of the animation through CNN and RNN, combines GNN and natural language processing algorithms to determine the animation effect, and evaluates the optimal solution through multiple synthesis algorithms to generate high Quality video frames and files.
It realizes automatic detection, precise analysis and flexible splitting of PPT animations, improves the automation level of video generation and the smoothness and authenticity of animation effects, simplifies the video production process, and lowers the threshold for video production.
Smart Images
Figure CN119131658A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of video synthesis and artificial intelligence content generation, and in particular to a method and system for automatic animation recognition and splitting of PPT synthesized videos. Background Art
[0002] Existing technologies have certain limitations when processing complex animation effects in PPT, especially when animation effects are nested or used in combination, identification and separation become particularly difficult. This difficulty not only affects the smoothness and authenticity of the animation, but also the low degree of automation when converting PPT animation effects to videos, resulting in low efficiency in video generation. In addition, during the conversion process, the animation effects may be distorted, especially in terms of animation speed and timeline synchronization, which seriously affects the quality of the final video. Summary of the invention
[0003] The purpose of the present invention is to provide a method and system for automatic animation recognition and splitting of PPT composite videos, so as to solve the above-mentioned problems existing in the prior art.
[0004] In order to achieve the above object, the technical solution adopted by the present invention is as follows:
[0005] A method for automatic animation recognition and splitting of PPT composite video, comprising the following steps:
[0006] S1. File parsing: Read the slide file, parse the internal structure and content of the slide file based on the pre-trained deep learning large language model, and obtain the content semantics of the slide;
[0007] S2, semantic parameter information extraction: traverse each slide and extract the specific parameter information of the slide content elements; by screening the dynamic content, analyze the relevant attributes of the animation elements, use CNN and RNN to extract the visual features and timing features of the animation respectively, aggregate and classify the visual features and timing features, and obtain and store the classified animation element information;
[0008] S3, animation production decision: Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides, and the video synthesis effects of multiple synthesis algorithms are evaluated to match the optimal synthesis algorithm as the final synthesis solution;
[0009] S4, video frame generation optimization: convert each slide into a video frame, use the classified animation element information and the final synthesis scheme to assign the animation elements to specific video frames, generate key frames and tune them;
[0010] S5. Video synthesis optimization: Encode key frames in sequence and combine audio processing technology and large language models to generate high-quality video files.
[0011] Preferably, step S1 specifically includes the following contents:
[0012] S11. Use Open XML on the .NET platform to understand the slide file structure and call the SDK API interface to read the slide content;
[0013] S12, use slide to traverse the slide content, read the specific content, and extract the note text;
[0014] S13. Use a pre-trained deep learning large language model to encode the extracted note text and convert it into a series of vector representations; and based on the language knowledge and patterns learned by the model on large-scale text data, analyze and understand the vectors and identify keywords to obtain a deep semantic representation of the slides.
[0015] Preferably, in step S12, text and pictures are accessed through the TextBody and B lip elements, and animations are accessed through the CT_Animation element.
[0016] Preferably, step S2 specifically includes the following contents:
[0017] S21, traversing each slide, extracting each element in the content of each slide, and analyzing parameter information of the element; the parameter information includes size, style, color, and position;
[0018] S22, preliminarily screening the animation content to identify elements containing animation effects;
[0019] S23, parsing the relevant attributes of the animation effect element to obtain structured animation attribute data; the relevant attributes of the animation element include animation type, target element, timeline, effect, and trigger;
[0020] S24, first standardize and normalize the acquired structured animation attribute data, and then use CNN and RNN networks to extract visual features and temporal features of the animation effect attribute data respectively;
[0021] S25. Based on the result of feature extraction, the K-Means clustering algorithm is used to classify similar animation elements together;
[0022] S26. Storing the classified animation element information to prepare for subsequent rapid synthesis of materials.
[0023] Preferably, in step S24, when extracting features using CNN and RNN networks,
[0024] Through the connection and weighted summation fusion strategy, the back propagation algorithm is used to adjust the network parameters so that the extracted features can better represent the dynamic behavior of the animation.
[0025] Preferably, step S3 specifically includes the following contents:
[0026] S31, converting the animation set consisting of the content semantics of the slides extracted in step S1 and the classified animation element information obtained in step S2 into a graph structure and inputting it into the trained GNN; the trained GNN will predict the dependency between the animations and give corresponding execution order and coherence suggestions;
[0027] S32, parsing the animation description text using natural language processing related algorithms;
[0028] S33, determining the play order, duration, and delay time of the animation;
[0029] S34, evaluating the sound effect of the animation effect on the video synthesis according to the animation effect type;
[0030] S35. Use a variety of synthesis algorithms to simulate and analyze the effects of video synthesis, set evaluation indicators for evaluating the synthesis effect, quantify and qualitatively evaluate the effects of various synthesis algorithms based on the set evaluation indicators, and match the optimal synthesis algorithm as the final synthesis solution.
[0031] Preferably, step S4 specifically includes the following contents:
[0032] S41. planning the overall framework of video synthesis according to the animation effect analysis result;
[0033] S42, splitting the slide into multiple detailed key frames, and determining the number of video frames, frame rate and resolution parameters of each slide;
[0034] S43, in chronological order and in combination with the frame rate, the classified animation element information obtained in step S2 and the optimal synthesis effect obtained in step S3 are sequentially allocated to specific video frames to generate high-quality key frame images, and the generator network algorithm is used to learn the transition features and rules in the samples, and the intermediate transition frames are generated according to the information of the start and end frames;
[0035] S44. Use the discriminator network to evaluate the quality of the generated high-quality key frames, and combine the optimal strategy learned in the reinforcement learning algorithm to fine-tune the details of the key frames.
[0036] Preferably, the optimal strategy in step S44 includes a policy gradient algorithm Po l icy Grad ient and a deep deterministic policy gradient algorithm DDPG.
[0037] Preferably, step S5 specifically includes the following contents:
[0038] S51, using an efficient video encoding algorithm to encode the key frame obtained in step S4, and optimizing the compression and playback quality of the video file;
[0039] S52. Use the audio processing library to add background music and narration;
[0040] S53. Understand the video content through a large language model, extract the picture and audio features using a deep learning algorithm, optimize the parameter settings using a neural network, and tune the generated video; based on the expectations for video quality, clarify the specific goals of the tuning. If the evaluation results do not meet expectations, repeat the optimization process until the specific goals of the tuning are met;
[0041] S54. Provides an editing interface that allows you to customize parameters and preview optimization effects in real time.
[0042] The present invention also aims to provide a system for automatic animation recognition and splitting of PPT composite videos, which can implement the above-mentioned method, and the system comprises:
[0043] File parsing module: reads slide files, parses the internal structure and content of slide files based on the pre-trained deep learning large language model, and obtains the content semantics of the slides;
[0044] Semantic parameter information extraction module: traverses each slide and extracts the specific parameter information of the slide content elements; by screening the dynamic content and parsing the relevant attributes of the animation elements, CNN and RNN are used to extract the visual features and timing features of the animation respectively, and the visual features and timing features are aggregated and classified to obtain and store the classified animation element information;
[0045] Animation production decision module: Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides, and the video synthesis effects of multiple synthesis algorithms are evaluated to match the optimal synthesis algorithm as the final synthesis solution;
[0046] Video frame generation and optimization module: converts each slide into a video frame, uses the classified animation element information and the final synthesis scheme to assign animation elements to specific video frames, generates key frames and optimizes them;
[0047] Video synthesis optimization module: encodes key frames in sequence and combines audio processing technology and a large language model to generate high-quality video files.
[0048] The beneficial effects of the present invention are: 1. It solves the problem that it is difficult to automatically identify and accurately split the complex animation effects in PPT in the prior art. The present invention realizes the automatic detection, accurate analysis and flexible splitting of animations in the process of PPT to video conversion, which improves the automation level of video generation and the fluency and authenticity of animation effects. 2. It greatly simplifies the previously cumbersome and time-consuming video production process. Through the built-in advanced image recognition technology and animation analysis technology, it can intelligently analyze each page of content, font style, color matching and built-in animation effects in PPT, and automatically convert them into smooth and coherent video clips. Users do not need to have professional video editing skills, they only need to simply upload the PPT file and set basic parameters according to personal needs. The system can respond quickly and automatically complete the video synthesis work. It greatly reduces the threshold for video production, so that whether it is corporate publicity, education and training or personal sharing, you can easily have high-quality video content. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 is a flow chart of a method in an embodiment of the present invention;
[0050] Figure 2 is a flow chart of semantic parameter information extraction in an embodiment of the present invention;
[0051] Figure 3 is a flow chart of animation production decision making in an embodiment of the present invention;
[0052] Figure 4 is a flow chart of video frame generation optimization in an embodiment of the present invention;
[0053] Figure 5 It is a flow chart of video synthesis optimization in an embodiment of the present invention. DETAILED DESCRIPTION
[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.
[0055] like Figure 1 As shown, in this embodiment, a method for automatic animation recognition and splitting of PPT composite video is provided, including the following parts:
[0056] 1. File parsing
[0057] Read the slide file, parse the internal structure and content of the slide file based on the pre-trained deep learning large language model, and obtain the content semantics of the slide.
[0058] Specifically include the following contents:
[0059] 1.1. Use Open XML on the .NET platform to understand the slide file structure and call the SDK API interface to read the slide content.
[0060] 1.2. Use slider to traverse the slide content, read the specific content, and extract the note text. For example, text and pictures can be accessed through TextBody and B lip elements, and animations can be accessed through CT_Animation elements.
[0061] 1.3. Use a pre-trained deep learning large language model (such as GPT-4, Wenxin Yiyan, Tongyi Qianwen, etc.) to encode the extracted note text and convert it into a series of vector representations; and based on the language knowledge and patterns learned by the model on large-scale text data, analyze and understand the vectors and identify keywords to obtain a deep semantic representation of the slides.
[0062] 2. Semantic Parameter Information Extraction
[0063] Traverse each slide and extract the specific parameter information of the slide content elements; by screening the dynamic content and parsing the relevant attributes of the animation elements, use CNN and RNN to extract the visual features and timing features of the animation respectively, aggregate and classify the visual features and timing features, and obtain and store the classified animation element information.
[0064] like Figure 2 As shown, it specifically includes the following contents:
[0065] 2.1. Traverse each slide, extract each element in the content of each slide, and analyze the element's size, style, color, position and other parameter information.
[0066] 2.2. Conduct a preliminary screening of the animation content and identify the elements that contain animation effects.
[0067] 2.3. Analyze the relevant attributes of the animation effect elements, including animation type, target element, timeline, effect, trigger, etc., to obtain structured animation attribute data.
[0068] 2.4. The acquired structured animation attribute data is first standardized and normalized, and then the CNN and RNN networks are used to extract the visual features and timing features of the animation effect attribute data respectively. Through a simple connection and weighted summation fusion strategy, the back propagation algorithm is used to adjust the network parameters so that the extracted features can better represent the dynamic behavior of the animation, which is convenient for the subsequent accurate identification of different animation types and their details.
[0069] 2.5. Based on the results of feature extraction, use the K-Means clustering algorithm to classify similar animation elements together.
[0070] 2.6. Store the classified animation element information to prepare for the subsequent rapid synthesis of materials.
[0071] 3. Animation Production Decisions
[0072] Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides. By evaluating the video synthesis effects of multiple synthesis algorithms, the optimal synthesis algorithm is matched as the final synthesis solution.
[0073] like Figure 3 As shown, it specifically includes the following contents:
[0074] 3.1. Convert the animation set consisting of the content semantics of the slides extracted in the first part and the classified animation element information obtained in the second part into a graph structure and input it into the trained GNN; the trained GNN will predict the dependencies between animations and give corresponding execution order and coherence suggestions to ensure the logical order of animation execution and scene coherence.
[0075] 3.2. Use natural language processing related algorithms to parse the animation description text.
[0076] 3.3. Determine the animation playback order, duration, delay time and other parameters.
[0077] 3.4. Based on the type of animation effect, evaluate the sound effect of the video synthesis, such as whether special processing is required (such as fade-in, fade-out, transition effects).
[0078] 3.5. Use a variety of synthesis algorithms to simulate and analyze the effects of video synthesis, set evaluation indicators for evaluating the synthesis effect (such as visual fluency, sound effect adaptation, overall rhythm, resource consumption, etc.), and quantitatively and qualitatively evaluate the effects of various synthesis algorithms based on the set evaluation indicators, and match the optimal synthesis algorithm as the final synthesis solution.
[0079] 4. Video frame generation optimization
[0080] Each slide is converted into a video frame, and the classified animation element information and the final synthesis scheme are used to assign the animation elements to specific video frames, generate key frames and tune them.
[0081] like Figure 4 As shown, it specifically includes the following contents:
[0082] 4.1. Based on the results of animation effect analysis, plan the overall framework of video synthesis.
[0083] 4.2. Split the slides into multiple detailed key frames and determine the parameters such as the number of video frames, frame rate and resolution of each slide.
[0084] 4.3. According to the time sequence and frame rate, the classified animation element information obtained in step S2 and the optimal synthesis effect obtained in step S3 are allocated to specific video frames in sequence to generate high-quality key frame images. The generator network algorithm is used to learn the transition features and rules in the samples, and the intermediate transition frames are generated according to the information of the start and end frames.
[0085] 4.4. Use the discriminator network to evaluate the quality of the generated high-quality key frames. The evaluation may be carried out from aspects such as peak signal-to-noise ratio and structural similarity index. Combined with the optimal strategy learned in the reinforcement learning algorithm (such as policy gradient algorithm Policy Gradient and deep deterministic policy gradient algorithm DDPG), the key frame details are tuned to improve the naturalness of the overall animation effect.
[0086] 5. Video synthesis optimization
[0087] The key frames are encoded in sequence and combined with audio processing technology and a large language model to generate high-quality video files.
[0088] like Figure 5 As shown, it specifically includes the following contents:
[0089] 5.1. Use an efficient video coding algorithm (such as H.265 / HEVC) to encode the key frames obtained in the fourth part to optimize the compression and playback quality of the video file.
[0090] 5.2. Use audio processing libraries (such as FFmpeg's audio processing module) to add background music and narration to enhance the appeal of the video.
[0091] 5.3. Understand the video content through a large language model, extract the image and audio features using a deep learning algorithm, optimize the parameter settings using a neural network, and tune the generated video; based on the expectations for video quality, clarify the specific goals of the tuning (such as improving image clarity, enhancing audio fidelity, improving video fluency, etc.); if the evaluation results do not meet expectations, repeat the optimization process until the specific goals of the tuning are met;
[0092] 5.4. Provides an editing interface that allows you to customize parameters and preview optimization effects in real time.
[0093] In this embodiment, a system for automatic animation recognition and splitting of PPT composite videos is provided, and the system can implement the above-mentioned method, and the system includes:
[0094] (1) File parsing module: reads the slide file, parses the internal structure and content of the slide file based on the pre-trained deep learning large language model, and obtains the content semantics of the slide.
[0095] (2) Semantic parameter information extraction module: traverse each slide and extract the specific parameter information of the slide content elements; by screening the dynamic content and parsing the relevant attributes of the animation elements, use CNN and RNN to extract the visual features and timing features of the animation respectively, aggregate and classify the visual features and timing features, and obtain and store the classified animation element information.
[0096] (3) Animation production decision module: Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides. The video synthesis effects of multiple synthesis algorithms are evaluated and the optimal synthesis algorithm is matched as the final synthesis solution.
[0097] (4) Video frame generation and optimization module: Convert each slide into a video frame, use the classified animation element information and the final synthesis scheme to assign the animation elements to specific video frames, generate key frames and optimize them.
[0098] (5) Video synthesis optimization module: Encode key frames in sequence and combine audio processing technology and large language models to generate high-quality video files.
[0099] By adopting the above technical solution disclosed in the present invention, the following beneficial effects are obtained:
[0100] The present invention provides a method and system for automatic animation recognition and splitting of PPT composite videos, which solves the problem that it is difficult to automatically recognize and accurately split complex animation effects in PPT in the prior art. The present invention realizes automatic detection, accurate analysis and flexible splitting of animations in the process of PPT to video conversion, improves the automation level of video generation and the fluency and authenticity of animation effects. It greatly simplifies the previously cumbersome and time-consuming video production process. Through the built-in advanced image recognition technology and animation analysis technology, it can intelligently analyze the content, font style, color matching and built-in animation effects of each page in PPT, and automatically convert them into smooth and coherent video clips. Users do not need to have professional video editing skills, they only need to simply upload PPT files and set basic parameters according to personal needs, and the system can respond quickly and automatically complete the video synthesis work. It greatly reduces the threshold for video production, so that whether it is corporate publicity, education and training or personal sharing, you can easily have high-quality video content.
[0101] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention.
Claims
1. A method for automatic animation recognition and splitting of PPT composite video, characterized by: The following steps are included: S1. File parsing: Read the slide file, parse the internal structure and content of the slide file based on the pre-trained deep learning large language model, and obtain the content semantics of the slide; S2, semantic parameter information extraction: traverse each slide and extract the specific parameter information of the slide content elements; by screening the dynamic content, analyze the relevant attributes of the animation elements, use CNN and RNN to extract the visual features and timing features of the animation respectively, aggregate and classify the visual features and timing features, and obtain and store the classified animation element information; S3, animation production decision: Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides, and the video synthesis effects of multiple synthesis algorithms are evaluated to match the optimal synthesis algorithm as the final synthesis solution; S4, video frame generation optimization: convert each slide into a video frame, use the classified animation element information and the final synthesis scheme to assign the animation elements to specific video frames, generate key frames and tune them; S5. Video synthesis optimization: Encode key frames in sequence and combine audio processing technology and large language models to generate high-quality video files.
2. The method for automatic animation recognition and splitting of PPT composite video according to claim 1 is characterized in that: Step S1 specifically includes the following contents: S11. Use Open XML on the .NET platform to understand the slide file structure and call the SDK API interface to read the slide content; S12, use slide to traverse the slide content, read the specific content, and extract the note text; S13. Use a pre-trained deep learning large language model to encode the extracted note text and convert it into a series of vector representations; and based on the language knowledge and patterns learned by the model on large-scale text data, analyze and understand the vectors and identify keywords to obtain a deep semantic representation of the slides.
3. The method for automatic animation recognition and splitting of PPT composite video according to claim 2 is characterized in that: In step S12, text and pictures are accessed through the TextBody and Blip elements, and animations are accessed through the CT_Animation element.
4. The method for automatic animation recognition and splitting of PPT composite video according to claim 1, characterized in that: Step S2 specifically includes the following contents: S21, traversing each slide, extracting each element in the content of each slide, and analyzing parameter information of the element; the parameter information includes size, style, color, and position; S22, preliminarily screening the animation content to identify elements containing animation effects; S23, parsing the relevant attributes of the animation effect element to obtain structured animation attribute data; the relevant attributes of the animation element include animation type, target element, timeline, effect, and trigger; S24, first standardize and normalize the acquired structured animation attribute data, and then use CNN and RNN networks to extract visual features and temporal features of the animation effect attribute data respectively; S25. Based on the result of feature extraction, the K-Means clustering algorithm is used to classify similar animation elements together; S26. Storing the classified animation element information to prepare for subsequent rapid synthesis of materials.
5. The method for automatic animation recognition and splitting of PPT composite video according to claim 4 is characterized in that: In step S24, when extracting features using CNN and RNN networks, Through the connection and weighted summation fusion strategy, the back propagation algorithm is used to adjust the network parameters so that the extracted features can better represent the dynamic behavior of the animation.
6. The method for automatic animation recognition and splitting of PPT composite video according to claim 1 is characterized in that: Step S3 specifically includes the following contents: S31, converting the animation set consisting of the content semantics of the slides extracted in step S1 and the classified animation element information obtained in step S2 into a graph structure and inputting it into the trained GNN; the trained GNN will predict the dependency between the animations and give corresponding execution order and coherence suggestions; S32, parsing the animation description text using natural language processing related algorithms; S33, determining the play order, duration, and delay time of the animation; S34, evaluating the sound effect of the animation effect on the video synthesis according to the animation effect type; S35. Use a variety of synthesis algorithms to simulate and analyze the effects of video synthesis, set evaluation indicators for evaluating the synthesis effect, quantify and qualitatively evaluate the effects of various synthesis algorithms based on the set evaluation indicators, and match the optimal synthesis algorithm as the final synthesis solution.
7. The method for automatic animation recognition and splitting of PPT composite video according to claim 1 is characterized in that: Step S4 specifically includes the following contents: S41. planning the overall framework of video synthesis according to the animation effect analysis result; S42, splitting the slide into multiple detailed key frames, and determining the number of video frames, frame rate and resolution parameters of each slide; S43, in chronological order and in combination with the frame rate, the classified animation element information obtained in step S2 and the optimal synthesis effect obtained in step S3 are sequentially allocated to specific video frames to generate high-quality key frame images, and the generator network algorithm is used to learn the transition features and rules in the samples, and the intermediate transition frames are generated according to the information of the start and end frames; S44. Use the discriminator network to evaluate the quality of the generated high-quality key frames, and combine the optimal strategy learned in the reinforcement learning algorithm to fine-tune the details of the key frames.
8. The method for automatic animation recognition and splitting of PPT composite video according to claim 7 is characterized in that: The optimal strategy in step S44 includes the policy gradient algorithm Policy Gradient and the deep deterministic policy gradient algorithm DDPG.
9. The method for automatic animation recognition and splitting of PPT composite video according to claim 1, characterized in that: Step S5 specifically includes the following contents: S51, using an efficient video encoding algorithm to encode the key frame obtained in step S4, and optimizing the compression and playback quality of the video file; S52. Use the audio processing library to add background music and narration; S53. Understand the video content through a large language model, extract the picture and audio features using a deep learning algorithm, optimize the parameter settings using a neural network, and tune the generated video; based on the expectations for video quality, clarify the specific goals of the tuning. If the evaluation results do not meet expectations, repeat the optimization process until the specific goals of the tuning are met; S54. Provides an editing interface that allows you to customize parameters and preview optimization effects in real time.
10. A system for automatic animation recognition and splitting of PPT composite videos, characterized by: The system can implement the method described in any one of claims 1 to 9, and the system includes: File parsing module: reads slide files, parses the internal structure and content of slide files based on the pre-trained deep learning large language model, and obtains the content semantics of the slides; Semantic parameter information extraction module: traverses each slide and extracts the specific parameter information of the slide content elements; by screening the dynamic content and parsing the relevant attributes of the animation elements, CNN and RNN are used to extract the visual features and timing features of the animation respectively, and the visual features and timing features are aggregated and classified to obtain and store the classified animation element information; Animation production decision module: Based on the content semantics of the slides and the classified animation element information, the trained GNN network and natural language processing related algorithms are used to determine the animation effect of the slides, and the video synthesis effects of multiple synthesis algorithms are evaluated to match the optimal synthesis algorithm as the final synthesis solution; Video frame generation and optimization module: converts each slide into a video frame, uses the classified animation element information and the final synthesis scheme to assign animation elements to specific video frames, generates key frames and optimizes them; Video synthesis optimization module: encodes key frames in sequence and combines audio processing technology and a large language model to generate high-quality video files.