An end-to-cloud decoupled AI animation generation method and system

Abstract

This invention discloses an edge-cloud decoupled AI animation generation method and system, relating to the fields of AI content generation and cloud computing. This invention delegates all AI animation generation computing power to the cloud, with the terminal only handling instruction input and image display, without participating in core computation. The cloud tests the terminal hardware performance and generates an adapted instruction set to complete AI animation generation and adaptive image quality compensation, smoothing out differences in image quality across devices. Then, through SVC layered encoding for lightweighting, intelligent scheduling dynamically adapts network transmission to the terminal for decoding and playback. The system caches user creation preferences and employs weak network caching and breakpoint resume fault tolerance mechanisms to ensure smooth playback even on weak networks. This invention decouples animation quality from terminal hardware, allowing low-end and mid-range devices to use high-definition AI animation generation capabilities without upgrades, lowering the barrier to entry, ensuring strong cross-terminal display consistency, and adapting to application scenarios such as roadshows, product visualization, educational courseware, and patent animations.

Description

1. Technical Field

[0001] This invention relates to the fields of artificial intelligence content generation, cloud computing, lightweight video encoding, and edge-cloud collaborative scheduling. Specifically, it is an AI animation generation method and system that decouples the edge and cloud and offloads all computing power. It can stably generate high-quality demonstration-type AI animations on low-configuration mobile terminals and is suitable for scenarios such as startup roadshows, patent application visualization, product function demonstrations, e-commerce displays, and educational courseware animation generation. 2. Background Technology

[0002] Currently, mainstream AI animation generation solutions on the market generally adopt a traditional architecture that relies primarily on edge inference and is assisted by the cloud. In practical applications, this architecture suffers from the following inherent industry defects: 1) Animation generation quality is heavily tied to the performance of the local hardware on the terminal. Low-spec devices are prone to rendering stutters, blurry images, failed generation tasks, and software crashes; 2) The local computing power consumption on the edge is extremely high. Prolonged operation can easily cause device overheating and frequency reduction, background process crashes, and even system crashes; 3) The underlying logic relies on iterative upgrades of the terminal hardware to improve generation capabilities, significantly raising the barrier to entry for ordinary users and small startups; 4) Existing cloud-based animation solutions only perform simple image forwarding, lacking front-end terminal performance adaptation and end-to-end image quality closed-loop compensation, resulting in significant differences in animation display effects across different terminal configurations; 5) General video transmission strategies cannot adapt to the dynamic bitstream characteristics of animations. Transmission is prone to interruption, image tearing, and stuttering in weak network environments, leading to a poor overall user experience. This invention addresses the aforementioned industry pain points by employing a complete closed-loop architecture that integrates full computing power cloud offloading, dynamic weight pre-adaptation, end-to-end image quality compensation, SVC layered encoding, terminal-network collaborative intelligent scheduling, and long-term caching of user preferences. This architecture completely decouples the entire animation generation chain from terminal hardware performance, enabling mobile terminals of all price ranges to stably output standardized, high-quality AI animations without requiring hardware upgrades. 3. Summary of the Invention

[0003] 3.1 Technical problems to be solved 1) Achieve complete decoupling of the full computing power of AI animation generation from the local hardware of the terminal, and completely get rid of the underlying limitations of the terminal's CPU and GPU performance on animation generation capabilities; 2) By using a multi-dimensional performance adaptation and loop verification mechanism, the animation generation benchmark for terminals with different configurations is unified, eliminating the problem of image quality and smoothness fluctuations caused by hardware differences. 3) Scalable layered encoding is used to reduce the amount of data transmitted, adapt to complex and ever-changing network environments, and greatly improve the stability of transmission and playback in weak network scenarios; 4) Achieve smooth generation and display of high-quality AI animations across all categories of mobile terminals without modifying or upgrading terminal hardware; 5) Retain users' long-term creative preference data to enable pre-adaptation of parameters in subsequent generation processes, thereby improving interaction efficiency and the fit between animation needs. 3.2 Technical Solution

[0004] Referring to the system architecture shown in Figure 2, the complete technical chain of this invention is as follows: 1) Full computing power offloading in the cloud: All rendering, compositing and computing tasks of AI animation generation are carried out by the heterogeneous computing power processing module 11 of cloud server 1. Terminal device 2 does not participate in the core computing power operation, but only undertakes the functions of command sending and receiving and screen display. 2) Dynamic weight pre-adaptation: The dynamic weight performance adaptation module 12 of the cloud server 1 performs multi-dimensional weight testing and compliance loop verification on the terminal CPU, GPU, memory, screen parameters, and network bandwidth to generate a dedicated animation generation instruction set. 3) Instruction-driven animation generation: The animation generation module 13 generates customized AI animations based on the adapted exclusive instruction set. 4) Adaptive image quality closed-loop compensation: The image quality compensation module 14 dynamically adjusts the resolution, frame rate, and color depth according to the terminal decoding limit and screen display parameters, and completes image distortion correction and color balance optimization. 5) SVC Lightweight Layered Encoding: The SVC layered encoding module 15 performs scalable layered encoding on the finished animation, splitting the basic low-level bitstream and the high-definition enhanced bitstream to adapt to dynamic bandwidth changes. 6) End-to-cloud collaborative intelligent scheduling: The intelligent scheduling module 16 monitors the network status in real time, dynamically switches encoding layers, adjusts the transmission bit rate, and provides supporting caching, breakpoint resume, and network fluctuation retransmission fault tolerance mechanisms. 7) Long-term caching of user preferences: The user habit caching module 17 retains the user's historical command style and generation parameter preferences for the adaptation of pre-parameters in subsequent generation tasks. 8) Lightweight decoding and display on the terminal: The terminal device 2 sends out a request through the instruction input module 21, the video decoding module 22 completes lightweight decoding of the bitstream, and finally the animation display module 23 outputs the animation screen. 3.3 Beneficial Effects 1) The underlying animation generation capability is completely decoupled from the terminal hardware, so it can be adapted and used by all mid-to-low-end mobile devices without hardware upgrades; 2) The terminal has no core computing power load, completely solving the inherent pain points of the industry such as lag, overheating, crashes and freezes in traditional terminal solutions; 3) The dual closed-loop design of front-end adaptation and image quality compensation enables unified and standardized animation output across all terminals, resulting in extremely strong cross-terminal screen consistency; 4) SVC layered encoding combined with an intelligent network scheduling mechanism ensures stable and smooth playback even in weak network environments, significantly improving network fault tolerance and adaptability. 5) Built-in long-term caching mechanism for user habits, which significantly improves the efficiency of subsequent animation generation and continuously optimizes the creative interaction experience; 6) The overall solution has low deployment cost, wide applicability, strong scalability, and is very easy to achieve large-scale popularization and application. 4. Description of the attached figures Figure 1. Overall flowchart of the AI ​​animation generation method with edge-cloud decoupling of the present invention; Figure 2. Block diagram of the AI ​​animation generation system with edge-cloud decoupling of the present invention; Figure 3. Execution flowchart of the dynamic weight performance adaptation module of the present invention; Figure 4. Processing flowchart of the adaptive image quality compensation module of the present invention; Figure 5. Implementation flowchart of the edge-cloud collaborative intelligent scheduling module of the present invention. Figure labeling: 1— Cloud server; 11— Heterogeneous computing power processing module; 12— Dynamic weight performance adaptation module; 13— Animation generation module; 14— Image quality compensation module; 15— SVC layered encoding module; 16— Intelligent scheduling module; 17— User habit caching module; 2— Terminal device; 21— Command input module; 22— Video decoding module; 23— Animation display module; 3— Communication network. 5. Detailed Implementation 5.1 Example 1: Complete process of the present invention (corresponding to Figures 1 and 2) Based on the system architecture shown in Figure 2, the complete execution steps of this invention are as follows: 1) The user issues an AI animation generation request command through the command input module 21 of the terminal device 2; 2) The instruction is transmitted to the cloud server 1 via communication network 3; 3) The dynamic weight performance adaptation module 12 of cloud server 1 starts terminal performance benchmark testing and completes multi-dimensional parameter collection and weight adaptation verification. 4) The heterogeneous computing power processing module 11 bears the entire computing load and drives the animation generation module 13 to generate the original AI animation according to the adapted instruction set; 5) The image quality compensation module 14 performs adaptive image quality correction on the original animation to match the upper limit of the terminal's display and decoding capabilities; 6) The SVC layered encoding module 15 performs layered lightweight encoding on the compensated animation; 7) The intelligent scheduling module 16 monitors the network status in real time, dynamically adjusts the transmission code stream and layering strategy, and transmits data to the terminal through the communication network 3; 8) The video decoding module 22 of terminal device 2 completes lightweight decoding of the encoded bitstream; 9) The animation display module 23 completes the final animation playback and display; 10) After the entire process is completed, the user habit cache module 17 retains the instruction preferences and style parameters for subsequent pre-adaptation generation. 5.2 Example 2: Dynamic Weight Performance Adaptation Process (corresponding to Figures 2 and 3) The dynamic weight performance adaptation module 12 of cloud server 1 starts the adaptation process, as shown in Figure 3: 1) Send a parameter acquisition request to terminal device 2 to obtain the original parameters of terminal CPU computing power, GPU rendering capability, memory limit, screen resolution, color bit depth, and uplink network bandwidth; 2) Assign differentiated adaptation weights to each parameter to construct a terminal hardware capability evaluation model; 3) Performance compliance assessment: If the terminal performance does not meet the standard, dynamically adjust the generation strategy threshold and re-perform the adaptation parameter verification. If the terminal performance meets the requirements, a custom-matched animation generation instruction set will be generated; 4) Output the final adapted instruction set to the animation generation module 13. 5.3 Example 3 Adaptive Image Quality Compensation Process (corresponding to Figures 2 and 4) The image quality compensation module 14 of cloud server 1 executes a closed-loop compensation process, as shown in Figure 4: 1) Receive the original AI animation materials output by the animation generation module 13; 2) Retrieve the display parameters and decoding performance limit data of terminal device 2; 3) Dynamically adapt and adjust the animation output resolution, playback frame rate, and color depth according to differences in terminal hardware; 4) Perform edge correction, color balance, and motion distortion compensation optimization; 5) Output the animation file after image quality calibration and transfer it to the SVC layer encoding module 15. 5.4 Example 4: Edge-Cloud Collaborative Intelligent Scheduling Process (corresponding to Figures 2 and 5) The intelligent scheduling module 16 of cloud server 1 controls the entire link transmission process, as shown in Figure 5: 1) Receive the animation layered bitstream preprocessed by the SVC layered encoding module 15; 2) Activate the end-to-cloud collaborative intelligent scheduling transmission strategy to monitor the bandwidth, transmission delay, and data packet loss rate of communication network 3 in real time; 3) Complete the pre-adaptation status verification: If the adaptation is not completed, it will return to the dynamic weight performance adaptation module 12 for re-adaptation; if the adaptation is completed, it will enter the transmission stage. 4) When the network is stable, invoke the full layered bitstream of SVC to maintain the complete transmission of high-definition animation; 5) In the event of network fluctuations or weak network conditions, the high-definition enhanced layer is automatically trimmed, and only the basic underlying bitstream is transmitted. Data caching, breakpoint resumption, and packet loss retransmission fault tolerance mechanisms are activated simultaneously. 6) The transmission task ends when the terminal has fully received all the animation data. 5.5 Example 5: User Habit Caching Mechanism (corresponding to Figure 2) The user habit caching module 17 on cloud server 1 runs continuously and sustainably. 1) Collect and retain user's past command keywords, animation style preferences, aspect ratio parameters, and duration configuration data; 2) Establish a user-specific parameter database and complete encrypted data storage; 3) When new generation tasks are issued in the future, cached data will be automatically retrieved to complete the pre-filling of pre-parameters, reducing the time spent on the adaptation process and improving the efficiency of animation generation and its fit with requirements.

Claims

1. A decoupled AI animation generation method based on edge and cloud, characterized in that, Includes the following steps: All computing power tasks required for AI animation generation are offloaded to cloud server 1, and terminal device 2 only serves as an interactive interface for command input and animation display, and does not participate in core computing power operations; Cloud server 1 performs dynamic weighted performance adaptation on terminal device 2 according to preset performance benchmark test standards. After cyclic verification and compliance, it generates an animation generation instruction set that matches the terminal performance. The cloud server 1 generates AI animations according to the instruction set and performs adaptive image quality compensation processing to eliminate image quality problems caused by differences in terminal performance; The processed animation is then encoded using SVC (Scalable Layered Coding) technology for lightweight encoding. Through intelligent scheduling with edge-cloud collaboration, the transmission bitrate and hierarchical transmission strategy are dynamically adjusted according to network conditions. The data is transmitted to terminal device 2 via communication network 3 for decoding and display. At the same time, the user's historical generation habits are cached to optimize the subsequent generation process.

2. The method according to claim 1, characterized in that, The dynamic weighted performance adaptation includes multi-dimensional weight assignment tests on terminal CPU performance, GPU performance, memory size, screen display parameters, and network bandwidth. It has a built-in compliance loop verification mechanism, and automatically reverts to readjust the adaptation threshold when parameters do not match.

3. The method according to claim 1, characterized in that, The adaptive image quality compensation process includes dynamically adjusting the resolution, frame rate, and color depth of the animation based on the terminal's display and decoding capabilities, and simultaneously performing image distortion correction and color balance optimization.

4. The method according to claim 1, characterized in that, The lightweight encoding adopts the SVC scalable layered encoding standard, which supports the splitting of the high-definition enhancement layer and the basic bottom layer bitstream, and can adaptively complete bitstream trimming according to network bandwidth.

5. The method according to claim 1, characterized in that, The edge-cloud collaborative intelligent scheduling includes dynamically switching encoding layers and adjusting the transmission bitrate based on bandwidth and transmission delay, and activating data caching, breakpoint resumption, and packet loss retransmission fault tolerance mechanisms when the network fluctuates.

6. The method according to claim 1, characterized in that, It includes a user-generated habit caching step, which retains user history command preferences, animation style parameters, and aspect ratio configuration data for pre-adaptation of parameters in subsequent animation generation tasks.

7. An AI animation generation system with decoupled edge and cloud architecture, characterized in that, include: Cloud server 1 is used to perform all the computing power tasks for AI animation generation, including dynamic performance adaptation, animation generation, image quality compensation, SVC layered encoding, intelligent scheduling, and user habit caching. Terminal device 2 serves only as a user command input and animation display interface and does not participate in core computing power calculation; communication network 3 is used for bidirectional data interaction between cloud server 1 and terminal device 2. The cloud server 1 performs performance adaptation on the terminal device 2, generates a matching animation generation instruction set, generates an animation, and after image quality compensation and SVC layered encoding, transmits it to the terminal device 2 through the communication network 3 to complete decoding and display.

8. The system according to claim 7, characterized in that, The cloud server 1 specifically includes: a heterogeneous computing power processing module 11, used to carry out all AI animation generation heterogeneous computing power operation tasks; a dynamic weight performance adaptation module 12, used to perform multi-dimensional performance benchmark testing and weight rating verification on the terminal device 2, and generate a dedicated animation generation instruction set; an animation generation module 13, used to complete the entire AI animation generation process according to the adaptation instruction set; an image quality compensation module 14, used to perform adaptive image quality optimization and image correction for animation; an SVC layered encoding module 15, used to complete lightweight layered encoding compression of animation videos; an intelligent scheduling module 16, used to dynamically adjust the end-to-end transmission strategy according to network status; and a user habit caching module 17, used to store user historical generation preferences and style parameter data.

9. The system according to claim 7, characterized in that, The terminal device 2 specifically includes: an instruction input module 21, used to collect user animation generation requirement instructions; a video decoding module 22, used to receive the encoded bitstream transmitted from the cloud and complete lightweight decoding; and an animation display module 23, used for playing and visually displaying the final animation screen.

10. The system according to claim 7, characterized in that, The terminal device 2 includes at least one of smartphones, tablets, laptops, and smart wearable devices, and can be fully adapted and used without hardware upgrades.

Images