A human full-age-segment aigc native lightweight modeling system

Abstract

The application discloses a kind of human full age segment AIGC native lightweight modeling system, it is related to digital human modeling technical field. Including AIGC native lightweight modeling unit, model double-layer distillation compression unit, virtual scene fusion unit, digital personality cross-platform mutual recognition unit, modeling boundary control unit. Adopt AIGC native architecture to realize full age segment digital personality small sample multi-modal integrated modeling in combination with transfer learning, eliminate redundant calculation by feature layer and inference layer double-layer distillation compression, frame synchronization scheduling, dynamic computing power allocation mechanism is guaranteed to guarantee the stability of end-side rendering, power consumption is reduced by 75% compared with native model, end-side rendering frame rate is greater than or equal to 35fps and frame rate is not fluctuated;With the mutual recognition protocol of self-defined digital personality model realizes multi-system cross-platform mutual recognition;Limit human exclusive modeling and controllable hiding of non-human elements.The application effectively solves the problems of traditional digital human modeling, such as high dependence on computing power, poor end-side adaptation, and cross-platform incompatibility, and meets the needs of digital life popularization and landing.

Description

Technical Field

[0001] This invention relates to the field of digital human modeling technology, and in particular to a lightweight AIGC native modeling system for all ages of humans. Background Technology

[0002] Current digital human modeling and digital life construction technologies generally face challenges in adapting to civilian applications, lightweight implementation, and full lifecycle deployment.

[0003] Current digital human modeling technologies heavily rely on massive amounts of labeled data and high-intensity computing power, resulting in long modeling cycles and high costs, making it difficult to apply to home and consumer scenarios. Model files are also large, only able to run in the cloud, and cannot be smoothly deployed on mid-range and low-end edge devices such as mobile phones, tablets, and entry-level smart screens. Furthermore, existing modeling solutions often only replicate the human image, lacking natural integration with real-world scenarios like family life, leading to significant frame rate fluctuations and high latency in edge rendering, resulting in extremely poor compatibility with mid-range and low-end devices.

[0004] Furthermore, the industry lacks a unified cross-platform standard for digital personality models, resulting in incompatibility in model invocation, authentication, and access control between different systems, and significant data and model conversion losses. More critically, existing technologies have not been optimized for modeling all age groups of humans, have not developed a lightweight AIGC native modeling process, and have not adopted double-layer distillation compression to significantly reduce computing power, thus failing to meet the modeling and presentation needs of the entire life cycle of human digital life, as well as for edge-based and civilian applications.

[0005] Therefore, a lightweight AIGC native modeling system for all ages of humans is needed to solve problems such as long modeling cycle, high computing power consumption, large model size, poor edge adaptation, insufficient scene integration, and cross-platform incompatibility. Therefore, a lightweight AIGC native modeling system for all human age groups is proposed to address the above problems. Summary of the Invention

[0006] Purpose of the invention: This invention provides a lightweight AIGC native modeling system for all ages of humans. It solves the technical problems of traditional modeling, such as high computational power dependence, poor edge-side adaptation, unstable rendering, cross-platform incompatibility, weak scene fusion, and lack of boundary control, through five core technologies: integrated multimodal AIGC lightweight modeling for all ages, dual-layer progressive distillation and compression of feature layer and inference layer, lightweight fusion rendering of core family scenes, custom digital personality model mutual recognition protocol (specifically DPMMPV1.0), and human-specific modeling boundary control.

[0007] Technical Solution: To solve the above-mentioned technical problems, according to one aspect of the present invention, more specifically, it is a human all-age AIGC native lightweight modeling system, including an AIGC native lightweight modeling unit, a model two-layer distillation and compression unit, a virtual scene fusion unit, a digital personality cross-platform mutual recognition unit, and a modeling boundary control unit. Each unit works in concert to complete the entire process of digital personality execution from feature input, modeling, compression, scene fusion, cross-platform mutual recognition to boundary control.

[0008] 1. AIGC Native Lightweight Modeling Unit Based on acquired multimodal feature data of human faces, limbs, and voice across all ages, a lightweight AIGC generation model and transfer learning algorithm are integrated to complete the unified modeling of human faces, limbs, and voice. After model parameter optimization, it is adapted for independent deployment on mobile devices, and the model output format meets the standardized adaptation requirements of mobile device rendering, interaction, and consistent expression. After parameter optimization, the size of a single personality model is ≤448M, enabling rapid construction of small samples of digital personalities across all ages. The entire process of modeling a single personality takes ≤1.75 hours, and the model fidelity is ≥97.1%.

[0009] 2. Model double-layer distillation compression unit This unit employs a two-layer progressive compression process: feature layer distillation followed by inference layer distillation. It addresses the redundant computations and parameter bloat of the native AIGC digital personality model through targeted optimization. The specific execution process is as follows: Characteristic layer distillation (1) Construct a multimodal feature weight map of face, limbs and voice for all age groups, quantify the feature contribution of each layer of the model feature extraction network, and select the core feature parameters with a representation degree ≥95%; (2) Perform targeted pruning on noisy features, duplicate features, and redundant computation nodes with a representation degree of <5% to remove invalid feature channels and convolution kernels; (3) All core representations required for the uniqueness of digital personality identity, integrity of actions, and clarity of speech are retained, and the amount of invalid computation in the feature extraction stage is reduced by 70%.

[0010] Inference layer distillation (1) Perform path topology analysis on the forward inference link of the model, merge repeated inference branches, delete empty loop logic, and simplify nonlinear activation functions; (2) Quantize the 32-bit floating-point parameters of the model into 16-bit fixed-point parameters to reduce the parameter storage size without losing rendering accuracy; (3) Optimize the edge-side inference scheduling logic to transform serial computing into edge-side GPU parallel computing, reducing the computational overhead of the inference stage by 80%.

[0011] Derivation of Two-Layer Collaborative Optimization and Computing Power Reduction Through a two-layer progressive distillation optimization of the feature layer and the inference layer, combined with hardware adaptation compilation and parallel computing scheduling, compared with the native AIGC digital personality model, the overall computing power consumption of the system is reduced by 75%, with no loss in rendering accuracy and functional integrity, while fully preserving the model's rendering accuracy and functional integrity; after optimization, the digital personality model's rendering accuracy and functional integrity are not lost, and the memory usage on the device side is reduced by 68.5% in tandem.

[0012] 3. Virtual Scene Fusion Unit A lightweight civilian scene model library is built around the core family life scenes of living room, courtyard and bedroom to achieve natural integration and rendering of digital personality and virtual scene. The synchronization rate of digital personality and scene rendering is ≥99.5% and the integration degree of digital personality and virtual scene is ≥99.6%. It supports scene adaptive lightweight lossless trimming for terminal devices, adapting to the scene presentation needs of small screen terminal devices, and the presentation effect of small screen terminal devices is lossless after trimming.

[0013] The implementation logic for edge-side rendering with a frame rate of ≥35fps, no frame rate fluctuations, and no stuttering or latency. The basic computing power guarantee dual-layer distillation reduces the model computing power consumption by 75%, and reduces the single-frame rendering computing load of low-end edge devices (quad-core A53 / octa-core A55) to within 25% of the hardware carrying capacity threshold, providing a computing power foundation for stable frame rate.

[0014] The frame synchronization rendering scheduling mechanism uses the digital personality action frame as the reference clock to achieve frequency synchronization between the action frame and the rendering frame. The rendering frame rate is fixed at 35fps, preventing frame rate jumps caused by the asynchrony between rendering and action.

[0015] The dynamic computing power priority allocation system monitors the CPU / GPU utilization rate on the device side in real time, and prioritizes 80% of the hardware computing power to the digital personality rendering. The virtual scene automatically downgrades the rendering accuracy according to the device performance to ensure that the core rendering tasks are not disturbed.

[0016] Local preloading and cloud-free model, scene, and motion data are all preloaded to the device's memory, eliminating cloud transmission and network jitter. Rendering calculations are performed entirely locally, with a single-frame rendering latency of <34ms, resulting in no stuttering and no transmission delay.

[0017] Frame rate fluctuation control locks the rendering pipeline and memory usage, achieving long-term stable and fluctuation-free rendering.

[0018] 4. Cross-platform recognition unit for digital personality Equipped with a custom digital personality model mutual recognition protocol, specifically DPMMPV1.0, it predefines the authentication process, permission level rules, and billing interface standards for cross-platform digital personality calls. It adopts a plug-in extension architecture to achieve cross-platform identity authentication, permission control, and standardized interface compatibility for digital personalities. The protocol is compatible with multiple operating systems such as Android, iOS, HarmonyOS, Windows, and Linux, enabling lossless cross-platform calls to digital personalities and seamless integration without adaptation delays.

[0019] 5. Modeling Boundary Control Units The system is limited to building only human-specific digital personalities and human life scene models, without any independent non-human modeling and rendering logic. Non-human visual elements are only used as scene embellishments and account for ≤1.5% of the total. Non-human visual elements are static vector graphics without independent AIGC modeling and rendering algorithms. The system is equipped with a full-link intelligent hiding interface for non-human visual elements, which can achieve one-click hiding. The hiding operation does not affect the accuracy of digital personality modeling, the integrity of scene rendering, or the fusion presentation effect. After one-click hiding, the rendering synchronization rate and fusion degree of digital personality and scene remain unchanged.

[0020] This system is compatible with mid-to-low-end edge devices such as budget smartphones and entry-level smart screens. It can achieve independent local rendering without cloud computing power, with an edge rendering frame rate of ≥35fps, continuous operation with no frame rate fluctuations, no stuttering, no rendering delay, and edge rendering latency of <34ms.

[0021] Beneficial effects: The lightweight and rapid modeling for all ages adopts the AIGC native architecture combined with transfer learning to achieve rapid construction of small samples of digital personalities for all ages. The entire process of modeling a single personality takes ≤1.75 hours, improving modeling efficiency. The model size is optimized to ≤448M, perfectly adapting to independent deployment on the edge, reducing the threshold for civilian use.

[0022] The dual-layer distillation technology optimizes performance by progressively distilling the feature layer and inference layer. It eliminates redundant calculations and simplifies the inference path, reducing computational power consumption by 75% and memory usage by 68.5% compared to the native model. While maintaining rendering accuracy, it significantly improves the efficiency of edge operation.

[0023] The stable and lossless rendering on the device is achieved through five mechanisms: synchronous frequency locking, dynamic computing power allocation, local preloading, redundant process shutdown, and rendering pipeline locking. This ensures a frame rate of ≥35fps, no frame rate fluctuation, and latency of <34ms. There is no stuttering, no latency, and no screen tearing on mid-to-low-end devices, and the rendering stability meets the long-term use needs of civilian scenarios.

[0024] With strong scene integration and adaptability, the lightweight library for core family scenes, synchronous rendering, and lossless trimming achieve a digital personality and scene integration degree of ≥99.6% and a synchronization rate of ≥99.5%. It adaptively adapts to all sizes of edge devices, and the scene presentation effect is lossless.

[0025] The cross-platform, fully compatible, and interoperable custom digital personality model mutual recognition protocol (DPMMPV1.0) covers five major operating systems. Its plug-in architecture enables lossless calling and seamless integration without adaptation delays, breaking down barriers to calling digital personalities across multiple devices and platforms. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the framework of the present invention. Detailed Implementation

[0027] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 1 As shown, a lightweight AIGC-native modeling system for all age groups of humans includes an AIGC-native lightweight modeling unit, a model two-layer distillation and compression unit, a virtual scene fusion unit, a digital personality cross-platform mutual recognition unit, and a modeling boundary control unit. These units work together to complete the entire process of digital personality development, from feature processing, modeling, compression, and scene fusion to cross-platform deployment and boundary control. The specific implementation steps are as follows: AIGC Native Lightweight Modeling Unit Implementation The system acquires multimodal feature data of human faces, limbs, and voice across all age groups, and initiates a lightweight AIGC generation model and transfer learning algorithm to complete multimodal integrated modeling. The model parameters are optimized and adapted for edge deployment, keeping the size of a single personality model ≤448M, ensuring that the entire modeling process takes ≤1.75 hours and the model fidelity is ≥97.1%. The model output format meets the standardized adaptation requirements for edge rendering, interaction, and consistent expression, and can be directly connected to subsequent modules.

[0029] Model double-layer distillation compression unit implementation The completed native digital personality model is compressed and optimized using a two-layer progressive process: feature layer distillation and inference layer distillation. Feature layer distillation involves constructing a multimodal feature weight map for all age groups, quantifying feature contribution, selecting core feature parameters with a representation degree of ≥95%, and pruning redundant features and nodes to reduce invalid computations in the feature extraction stage. Inference layer distillation involves topological analysis of the forward inference path, merging duplicate branches, and quantizing parameters from 32-bit floating-point to 16-bit fixed-point types, transforming serial computation into parallel computation on the edge GPU to reduce computational overhead in the inference stage. After two-layer collaborative optimization and edge hardware adaptation compilation, the model's overall computational power consumption is reduced by 75% and memory usage is reduced by 68.5% compared to the native model, while fully preserving rendering accuracy and functional integrity.

[0030] Implementation of virtual scene fusion unit It utilizes lightweight model libraries for core home scenes such as the living room, courtyard, and bedroom to achieve integrated rendering of digital personas and virtual scenes, ensuring a rendering synchronization rate of ≥99.5% and an integration degree of ≥99.6%. Stable rendering is achieved through five mechanisms: basic computing power guarantee, frame synchronization scheduling, dynamic computing power allocation, local preloading, and frame rate fluctuation control. This keeps the single-frame rendering load within the hardware threshold, locks the rendering frame rate, prioritizes computing power allocation, loads all data locally, and locks the rendering pipeline, ultimately achieving a frame rate of ≥35fps, no frame rate fluctuation, and rendering latency of <34ms. Lightweight lossless trimming is performed according to the size of the device to adapt to the display needs of small-screen devices, resulting in no lag and no transmission delay.

[0031] Implementation of cross-platform mutual recognition unit for digital personality It loads a custom digital personality model mutual recognition protocol (specifically DPMMPV1.0), completes cross-platform authentication, permission classification and billing interface adaptation according to predefined rules, and adopts a plug-in architecture for extended compatibility; it enables lossless calls to multiple systems such as Android, iOS, HarmonyOS, Windows and Linux, and has no adaptation delay when interfacing with external systems.

[0032] Modeling Boundary Control Unit Implementation The system generates only human digital personality and life scene models by default, and disables independent modeling and rendering logic for non-human subjects; it limits non-human visual elements to static vector graphics, with a proportion of ≤1.5%; it enables the full-link intelligent hiding interface, which supports one-click hiding of non-human elements without affecting the accuracy of digital personality modeling, scene rendering integrity, synchronization rate and integration.

[0033] This system is compatible with mid-to-low-end edge devices such as budget smartphones and entry-level smart screens. It operates entirely without cloud computing power, enabling independent local rendering and running, and meeting the long-term stable usage needs of civilian scenarios.

[0034] Preferred overall implementation process This invention adopts a standardized pipeline operation mode, executing the entire chain sequentially as follows: feature data input → full-age digital personality modeling → two-layer distillation compression optimization → virtual scene fusion rendering → cross-platform interoperability → modeling boundary control → independent operation on the terminal side. Data is seamlessly connected and units collaborate in each stage. The specific optimized process is as follows: Feature data input and preprocessing The system proactively acquires multimodal feature data of human faces, limbs, and voices across all age groups, and completes data cleaning, normalization, and format standardization conversion to provide a compliant input source for subsequent modeling.

[0035] Lightweight digital personality modeling for all ages The AIGC native lightweight modeling process is initiated. Based on the preprocessed feature data, the lightweight AIGC generation model and transfer learning algorithm are combined to complete the integrated generation of multimodal facial, limb, and voice features. Through parameter optimization, the model size is controlled to ≤448M, ensuring that the modeling time is ≤1.75 hours, the fidelity is ≥97.1%, and the output is a standard digital personality model that can be directly used for compression.

[0036] Bilayer distillation compression and hardware adaptation The original model is sequentially fed into the feature layer distillation and inference layer distillation modules to complete two-layer optimization, eliminating redundant features, simplifying the inference path, and completing parameter quantization and parallel computing conversion. After being compiled by the edge hardware instruction set, the computing power consumption is reduced by 75% and the memory usage is reduced by 68.5%, generating a lightweight running model adapted to low-end and mid-range devices.

[0037] Virtual scene blending and stable rendering The system calls upon the core family scene library to complete the synchronous fusion rendering of digital personality and virtual scene, ensuring that the synchronization rate and fusion degree of scene and personality actions meet the standards. Through mechanisms such as frame synchronization frequency locking, dynamic computing power allocation, and local preloading, it achieves a stable rendering effect with a frame rate of ≥35fps, no frame rate fluctuation, and latency of <35ms, and completes lossless scene trimming according to device size.

[0038] Cross-platform interoperability and interface integration Based on the DPMMPV1.0 protocol, cross-platform identity authentication, permission classification and interface adaptation of digital personality are completed. Through the plug-in architecture, it is compatible with multiple operating systems and realizes lossless and delayless calling of digital personality between different terminal devices.

[0039] Modeling boundary compliance management The system automatically defines human-specific modeling rules and treats non-human elements as low-proportion static auxiliary elements; users can control non-human visual elements by hiding the interface with one click, ensuring the purity and compliance of the digital personality presentation.

[0040] Independent deployment and long-term operation at the edge The optimized digital personality model is deployed to devices such as budget smartphones and entry-level smart screens, achieving independent local rendering without cloud computing power. It runs smoothly for extended periods without lag or frame rate fluctuations, meeting the needs of normal civilian use.

[0041] Example 1 Based on the acquired multimodal feature data of human faces, limbs, and voice across all age groups, including the feature data of the user's parents across all age groups, the system of this invention is used to complete the construction of a digital personality and its implementation on the family side. The specific implementation process is as follows: Digital personality modeling: The system acquires multimodal feature data of the user's parents' face, body and voice, and completes integrated modeling of dual digital personalities in 1.7 hours through AIGC's native lightweight modeling process. The model sizes are 415M and 420M respectively, with a fidelity of 97.2%. The output format meets the standardized adaptation requirements of edge rendering, interaction and consistent expression.

[0042] Two-layer distillation compression: The model is optimized by two-layer progressive optimization, namely feature layer distillation and inference layer distillation. The feature layer removes redundant features to reduce invalid computation, and the inference layer reduces computational overhead through parameter quantization and parallelization. Through phased computational power weighting and edge compilation co-optimization, the overall computational power consumption is reduced by 75% and memory usage is reduced by 68.5% compared with the original model, with no loss of rendering accuracy.

[0043] Scene fusion and stable rendering: The living room lightweight scene library is used to complete the fusion rendering, with a synchronization rate of 99.6% and a fusion degree of 99.7%. After lossless trimming and adaptation to entry-level smart screens, the frame rate is 38fps with a fluctuation of 0.7fps and a rendering latency of 31ms through a stable rendering mechanism. It runs continuously for 30 minutes without any lag or delay.

[0044] Cross-platform interoperability: Digital personality authentication is completed through the DPMMPV1.0 protocol, which is compatible with HarmonyOS and Android systems, enabling seamless cross-platform access between smart screens and mobile devices.

[0045] Boundary control execution: The pet static vector graphic, which accounts for 1.2% of the scene, has no independent modeling and rendering logic; after being hidden with one click through the smart interface, the presentation effect of the digital personality and the scene remains unchanged.

[0046] In this embodiment, the model was successfully deployed on a mid-to-low-end smart screen, running independently locally without cloud computing power. It can smoothly perform everyday actions such as reading newspapers, drinking tea, and selecting vegetables, fully meeting the needs of civilian and scenario-based presentation.

[0047] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A lightweight AIGC-native modeling system for all ages of humans, comprising an AIGC-native lightweight modeling unit, a model bi-layer distillation and compression unit, a virtual scene fusion unit, a cross-platform digital personality recognition unit, and a modeling boundary control unit, characterized in that: The AIGC native lightweight modeling unit integrates a lightweight AIGC generation model and transfer learning algorithm based on the acquired multimodal feature data of human face, limbs and voice across all ages, to complete the integrated modeling of human face, limbs and voice. After optimization of model parameters, it is adapted for independent deployment on the edge, and the model output format meets the standardized adaptation requirements of edge rendering, interaction and consistent expression. The model's dual-layer distillation compression unit employs a dual-layer progressive optimization mechanism of feature layer distillation and inference layer distillation. Feature layer distillation preserves the core facial, limb, and voice representations of the digital personality across all age groups, while eliminating redundant features and noise parameters. Inference layer distillation simplifies the model's forward inference path, merges duplicate inference branches, and completes model parameter quantization and edge-side parallel computing optimization. Compared to the uncompressed native AIGC digital personality model, the overall computing power consumption is reduced by 75%, effectively reducing model computing power consumption and ensuring edge-side rendering stability. The virtual scene fusion unit builds a lightweight civilian scene model library around the core family life scenes of living room, courtyard and bedroom, realizes the natural fusion rendering of digital personality and virtual scene, the synchronization rate of digital personality and scene rendering is ≥99.5%, supports scene adaptive lightweight lossless trimming for terminal devices, and adapts to the scene presentation needs of small screen terminal devices. The digital personality cross-platform mutual recognition unit is equipped with a custom digital personality model mutual recognition protocol to achieve cross-platform identity authentication, access control and interface standardization compatibility of digital personality; The modeling boundary control unit limits the system to only construct human-specific digital personality and human life scene models, without any independent non-human modeling and rendering logic. Non-human visual elements are only used as scene embellishments and account for ≤1.5%. It is equipped with a full-link intelligent hiding interface for non-human visual elements, which can realize one-click hiding operation. The hiding operation does not affect the accuracy of digital personality modeling, the integrity of scene rendering, and the fusion presentation effect.

2. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, After parameter optimization, the AIGC native lightweight modeling unit has a single personality model size of ≤448M, which can realize the rapid construction of small samples of digital personality across all ages. The entire process of single personality modeling takes ≤1.75 hours, and the model fidelity is ≥97.1%.

3. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The feature layer distillation of the dual-layer distillation compression unit of the model reduces the amount of invalid computation in the feature extraction stage by 70% by constructing a multimodal feature weight map of the entire age range and performing feature contribution quantification on the feature extraction network. It retains only the core feature parameters with a representation degree of ≥95% and performs targeted pruning on redundant features and computation nodes with a representation degree of <5%.

4. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The inference layer distillation of the model's dual-layer distillation compression unit reduces the computational overhead of the inference stage by 80% through topology analysis of the model's forward inference link execution path, quantization of 32-bit floating-point parameters into 16-bit fixed-point parameters, and conversion of serial computation into parallel computation on the edge GPU.

5. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The virtual scene fusion unit uses an action frame-rendering frame synchronization frequency locking mechanism and a terminal-side dynamic computing power priority allocation mechanism to prioritize 80% of the hardware computing power for digital personality rendering. Combined with local preloading of model, scene, and action data on the terminal side, it achieves a stable terminal-side rendering frame rate of ≥35fps, with no frame rate fluctuations, no stuttering, and no transmission delay.

6. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The digital personality cross-platform mutual recognition unit is equipped with the digital personality model mutual recognition protocol DPMMPV1.0, which predefines the authentication process, permission level rules and billing interface standards for cross-platform digital personality calls. It adopts a plug-in extension architecture, is compatible with multiple operating systems such as Android, iOS, HarmonyOS, Windows and Linux, and realizes lossless cross-platform digital personality calls and seamless integration without adaptation delays.

7. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The non-human visual elements of the modeling boundary control unit are static vector graphics, without independent AIGC modeling and rendering algorithms. After one-click hiding, the rendering synchronization rate and integration degree of the digital personality and the scene remain unchanged.

8. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The digital personality of the virtual scene fusion unit has a fusion degree of ≥99.6% with the virtual scene, and the scene is presented on the small screen device without loss after lightweight and lossless trimming.

9. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, After the optimization of the model's dual-layer distillation compression unit, the rendering accuracy and functional integrity of the digital personality model are maintained without loss, while the memory usage on the client side is reduced by 68.5%.

10. The AIGC native lightweight modeling system for all age groups of humans according to claim 1, characterized in that, The AIGC native lightweight modeling unit, model two-layer distillation and compression unit, virtual scene fusion unit, digital personality cross-platform mutual recognition unit, and modeling boundary control unit work together to complete the entire process of digital personality execution from feature input, modeling, compression, scene fusion, cross-platform mutual recognition to boundary control.

Images