Three-dimensional animation effect simulation device and method based on AIGC

Abstract

The invention discloses a three-dimensional animation effect simulation device and method based on AIGC in the technical field of artificial intelligence and virtual simulation, and the device comprises an AIGC generation module which is used for generating a three-dimensional model and an animation sequence based on an input instruction; the physical engine verification module is used for verifying the physical rationality of the three-dimensional model in real time based on the animation sequence, including collision detection, rigid body dynamics and centroid stability calculation; the computing power dynamic allocation module is used for dynamically allocating GPU resources to a rendering or physical simulation task according to scene requirements; and the cross-scene adaptation module is used for calling a preset physical parameter template library and adapting different types of scenes generated by the AIGC. According to the method, through deep coupling of the AI and the physical engine, heterogeneous computing power collaboration and dynamic LOD optimization, the problems of low efficiency, physical distortion, resource contention and the like in traditional animation production are solved, and the interaction authenticity and the computing efficiency are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and virtual simulation, and in particular relates to a three-dimensional animation effect simulation device and method based on AIGC. Background Art

[0002] In traditional 3D animation production, model building, animation generation, and physical effect verification rely heavily on manual labor, resulting in low efficiency, high costs, and a lack of physical realism. While recent progress has been made in 3D modeling and animation generation using artificial intelligence generated content (AIGC) technologies (such as GANs and diffusion models), the resulting models often lack physical constraints, resulting in distorted motion (such as character sliding and object penetration) or illogical dynamics (such as excessive joint torque). This requires extensive manual correction, requiring manual adjustment of collision bodies and dynamic parameters using tools like Maya and Blender, a process that can take hours to days.

[0003] At the physics simulation level, existing technologies (such as NVIDIA PhysX and Unity DOTS) can achieve rigid body collision and fluid dynamics calculations, but their real-time collaboration capabilities with AIGC-generated content are insufficient. Specifically, 1) physical parameters (such as mass and friction coefficient) need to be manually preset and cannot adapt to the diverse models dynamically generated by AIGC; 2) high-precision physics simulation and rendering tasks compete for GPU resources, resulting in frame rate fluctuations (±15fps) and delays (>20ms) in real-time interactive scenarios, affecting the user experience. For example, in game development, physics calculations for complex scenes require more than 60% of GPU resources, forcing developers to reduce rendering accuracy to maintain performance. In addition, cross-scene physics adaptation issues are prominent. The requirements for physical rules in fields such as film and television, games, and industrial simulation vary significantly (for example, science fiction scenes require zero-gravity simulation, while real-life scenes require precise rigid body collisions). Traditional solutions rely on manually configured parameter templates, and switching is time-consuming and error-prone. At the same time, existing technologies have also attempted to integrate AI and physical simulation, but they focus on offline training environments and lack real-time interaction capabilities; dynamic detail level technology is mostly aimed at rendering optimization, ignoring the dynamic regulation of physical simulation accuracy.

[0004] Therefore, it is necessary to propose a 3D animation effect simulation device and method based on AIGC, which can deeply integrate AIGC generation, real-time physical verification, dynamic computing power allocation and cross-scene adaptation to solve the above problems. Summary of the Invention

[0005] In order to solve the above problems, the purpose of the present invention is to provide a three-dimensional animation effect simulation device and method based on AIGC, which solves the problems of low efficiency, physical distortion, resource contention, etc. in traditional animation production through deep coupling of AI and physics engine, heterogeneous computing power collaboration and dynamic LOD optimization, and significantly improves the interactive realism and computing efficiency.

[0006] In order to achieve the above-mentioned object, the technical solution of the present invention is as follows: a 3D animation effect simulation device based on AIGC, comprising: an AIGC generation module for generating a 3D model and an animation sequence based on input instructions;

[0007] The physics engine verification module is used to verify the physical rationality of 3D models in real time based on animation sequences, including collision detection, rigid body dynamics, and center of mass stability calculations;

[0008] The computing power dynamic allocation module is used to dynamically allocate GPU resources to rendering or physical simulation tasks according to scene requirements;

[0009] The cross-scene adaptation module is used to call the preset physical parameter template library to adapt to different types of scenes generated by AIGC, and seamlessly stitch the three-dimensional model generated by the AIGC generation module with the animation sequences of different scenes.

[0010] Furthermore, during the training phase of the AIGC generation module, the output model of the AIGC generation module adopts a physical constraint training model;

[0011] The physical constraint training model includes: center of mass projection error loss unit, collision volume penetration loss unit and joint torque rationality loss unit;

[0012] The centroid projection error loss unit is used to constrain the three-dimensional model's motion to conform to static equilibrium;

[0013] The collision volume penetration loss unit is used to block non-physical overlap between 3D models;

[0014] The joint torque rationality loss unit is used to constrain the limb motion of the three-dimensional model to conform to biomechanics.

[0015] Furthermore, the dynamic computing power allocation module divides the GPU computing cycle through hardware-level time slice rotation, and divides the computing unit into physically isolated rendering partitions and physical partitions based on the GPU hardware scheduler. Each partition exclusively uses the SM cluster and L2 cache resources, and creates an independent task queue through CUDA Stream.

[0016] Furthermore, the AIGC generation module collaborates with the physics engine verification module to perform dynamic level of detail adjustment;

[0017] Dynamic Level of Detail adjustments include:

[0018] Reduce the mesh accuracy of the 3D model away from the collision area to 10%-30% of the original 3D model;

[0019] For model parts in animation sequences where the frequency of physical interaction between the 3D model and user operations or the environment exceeds a preset threshold, the full-precision mesh is retained and the physical calculation frequency is increased to 120Hz.

[0020] Furthermore, the cross-scenario adaptation module is also used to implement physical parameter migration through keyword-driven template switching; the specific steps are as follows:

[0021] a. Parse the scene keywords of the input text in the AIGC generation module;

[0022] b. Match the physical parameter set in the preset template library;

[0023] c completes parameter loading within 100ms and seamlessly integrates with the current animation sequence.

[0024] Furthermore, the physics engine module is also used to complete the following operations within a single frame period: detecting physical conflicts of the three-dimensional model; triggering the AIGC generation module to perform local parameter correction on the conflict area;

[0025] Correction methods include: limb rotation angle adjustment; collision volume expansion compensation.

[0026] Furthermore, the dynamic computing power allocation module adopts a collaborative architecture of heterogeneous computing units to automatically switch computing units according to the rendering / physics task type.

[0027] Furthermore, the physical engine verification module is also used to generate multi-resolution physical meshes for the three-dimensional model;

[0028] Multi-resolution physical grids include: high-resolution collision volume, grid density ≥ 1000 faces / cubic decimeter; low-resolution proxy volume, grid density ≤ 100 faces / cubic decimeter;

[0029] The motion state of the proxy body and the high-resolution mesh is synchronized through a deformation mapping algorithm.

[0030] Furthermore, in the computing power dynamic allocation module, when the GPU utilization rate is ≥95%, the hierarchical degradation strategy is initiated;

[0031] Tiered downgrade strategies include:

[0032] Level 1 downgrade: turn off ray tracing reflections and global illumination;

[0033] Second level degradation: switches the physics simulation accuracy to low-resolution proxy mode;

[0034] Level 3 degradation: Freeze the physics calculations of the non-interactive background model.

[0035] Furthermore, a 3D animation effect simulation method based on AIGC has the following specific steps: Step 1, AIGC generation and physical constraint training, generating a 3D model and animation sequence through the AIGC generation module based on input instructions, and during the generation process, optimizing the output results in real time through the physical constraint training model;

[0036] In step 2, the physics engine verification module checks the physical rationality of the animation sequence frame by frame; based on the complexity of the scene, tasks are dynamically allocated through the collaborative architecture of heterogeneous computing units;

[0037] Step 3: Cross-scenario parameter migration and local correction: parse the scene keywords in the AIGC input, match the physical parameter set from the preset template library, and perform local parameter correction in areas where physical conflicts are detected.

[0038] Step 4: Multi-resolution physical mesh and dynamic LOD optimization. Generate a multi-resolution physical network for the 3D model, and generate high-resolution collision bodies or low-resolution proxy bodies based on the physical interaction frequency and scene requirements. In the animation sequence, the model parts where the frequency of physical interaction between the 3D model and user operations or the environment exceeds the preset threshold will retain the full-precision mesh and increase the physical calculation frequency to 120Hz. The mesh accuracy of the 3D model far away from the collision area will be reduced to 10%-30% of the original 3D model.

[0039] The beneficial effects of adopting this solution are as follows: 1. In the traditional animation production process, the model generated by AIGC requires manual frame-by-frame adjustment of physical parameters (such as collision body and mass distribution), which takes several hours to several days. This solution embeds physical rules into the AIGC generation stage through physical constraint pre-training model and real-time verification-correction closed loop. For example: the center of mass projection error loss function is introduced during AI training to force the generated character movements to conform to static balance (such as the center of mass projection is always located on the supporting foot when walking), eliminating the "sliding" distortion problem; the physics engine verifies the collision volume penetration frame by frame, and provides real-time feedback to the AIGC module through local parameter correction (such as expanding the collision body 1.05 times), avoiding manual rework in traditional solutions.

[0040] 2. In traditional GPU architectures, rendering and physics computing compete for resources, resulting in frame rate fluctuations (±15fps) and latency (>20ms) in high-load scenarios. This solution achieves a performance leap through a collaborative architecture of heterogeneous computing units and a dynamic computing power allocation strategy. This offloads physics simulation tasks to the dynamic computing power allocation module, freeing up GPU resources for ray tracing and AI generation, significantly improving GPU utilization. Furthermore, through hardware-level time slice rotation (such as dynamically adjusting the physics time slot ratio to 60%-75%), physics calculation accuracy is improved while maintaining a stable high frame rate in highly dynamic scenes such as explosion special effects. A graded degradation strategy (disabling non-critical rendering effects → switching to proxy mode → freezing background physics) ensures system stability under extreme loads and avoids lag or crashes.

[0041] 3. This solution achieves a dual breakthrough in flexibility and authenticity through keyword-driven template migration and dynamic LOD optimization. It also supports dynamic adaptation and refined interaction across scenarios, expanding the application scenarios and user experience ceiling. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of an embodiment of the three-dimensional animation effect simulation device based on AIGC of the present invention.

[0043] Figure 2 Schematic diagram of an embodiment of the AIGC-based 3D animation effect simulation method of the present invention. DETAILED DESCRIPTION

[0044] The following is further described in detail through specific implementation methods:

[0045] Example 1:

[0046] Basically as attached Figure 1 Shown: A three-dimensional animation effect simulation device based on AIGC, comprising:

[0047] The AIGC generation module is used to generate a 3D model and animation sequence based on input instructions. During the training phase of the AIGC generation module, the output model of the AIGC generation module adopts a physical constraint training model, wherein the physical constraint training model includes: a center of mass projection error loss unit, a collision volume penetration loss unit, and a joint torque rationality loss unit;

[0048] The center of mass projection error loss unit is used to constrain the 3D model movement to conform to static equilibrium. The center of mass projection error loss (L_c) is calculated as follows:

[0049] Among them, c t is the coordinate of the center of mass of the character in the tth frame, P t is the geometric center of the support surface (such as the foot contact area), Proj支撑面 The center of mass is the vertical projection of the support surface. Its constraint goal is to force the trajectory of the center of mass during character actions (such as walking and jumping) to always be within the projection of the support surface to prevent sliding or imbalance.

[0050] The collision volume penetration loss unit is used to block non-physical overlap between 3D models. The calculation formula of the collision volume penetration loss (L_p) is as follows:

[0051]

[0052] Among them, d penetration (M i ,M j ) is the model M i With M j The penetration depth, ω ij The penetration penalty weight (e.g., key part penetration weight = 10, non-key part penetration weight = 1) is used to detect non-physical overlaps between models (e.g., a hand penetrating a wall) and adjust the topology of the AI-generated model by returning the gradient (e.g., shrinking the collision volume).

[0053] The joint torque rationality loss unit is used to constrain the limb motion of the three-dimensional model to conform to biomechanics. The calculation formula of the joint torque rationality loss (L_t) is as follows:

[0054]

[0055] Its constraint goal is to constrain skeletal movements (such as arm swinging and leg bending) to comply with biomechanical limits and prevent joints from over-rotating or torque overload.

[0056] The training phase of the AIGC generation module is as follows: 1. Data preparation: Pre-training is performed using a 3D motion dataset containing physical annotations (e.g., CMU MoCap with physical parameter labels); 2. Optimization algorithm: Adopting the AdamW optimizer, dynamically adjusting the loss weights (λc = 1.0, λp = 2.0, λt = 0.5); 3. Backpropagation: The physical loss gradient is jointly propagated back along with the generation model loss (e.g., adversarial loss and reconstruction loss) to achieve coordinated optimization of generation and physical plausibility. As a result, when generating character walking animations, the center of mass projection error is reduced from ±5cm in traditional solutions to ±0.3cm, while the inter-3D model penetration rate is reduced from 15% to approximately 0.5%.

[0057] The physics engine verification module verifies the physical plausibility of 3D models in real time based on animation sequences, including collision detection, rigid body dynamics, and center of mass stability calculations. Within a single frame, the physics engine module also performs the following operations: detecting physical collisions within the 3D model (such as penetration and center of mass offset); triggering the AIGC generation module to perform local parameter corrections in the conflicting area. These corrections include: 1. limb rotation angle adjustment within a ±5° range; and 2. collision volume expansion compensation (expansion coefficient 1.05 to 1.2 times).

[0058] The physics engine verification module also generates a multi-resolution physical mesh for the 3D model. This multi-resolution physical mesh consists of a high-resolution collision volume with a mesh density of ≥1000 faces / dm³ and a low-resolution proxy volume with a mesh density of ≤100 faces / dm³. The low-resolution proxy volume and the high-resolution mesh are synchronized using a deformation mapping algorithm.

[0059] Furthermore, the AIGC generation module and the physics engine verification module collaborate to perform dynamic level of detail adjustments. These adjustments include: 1. Reducing the mesh accuracy of 3D models away from collision zones to 10%-30% of the original 3D model; 2. Preserving the full-precision mesh in high-frequency interaction areas during animation sequences and increasing the physics calculation frequency to 120Hz.

[0060] The dynamic computing power allocation module is used to dynamically allocate GPU resources to rendering or physical simulation tasks based on scene requirements. This module divides the GPU computing cycle through hardware-level time slicing and, based on the GPU hardware scheduler, divides the computing unit into physically isolated rendering and physical partitions. Each partition has exclusive access to SM cluster and L2 cache resources, and creates an independent task queue through CUDA Stream.

[0061] Specifically, the asynchronous execution capability of CUDA Stream is utilized to create an independent task queue, and the computing unit is then divided into physically isolated partitions through the GPU hardware scheduler (such as the Multi-Instance GPU, MIG, of the Ampere architecture). This is then divided into rendering slots (Rendering Slot) and physical slots (Physics Slot) with a period of 16ms (corresponding to 60fps). For example, in highly dynamic scenes, the physical slot is allocated 10ms + the rendering slot is 6ms. The time slot ratio is then dynamically adjusted through hardware registers based on the real-time load (such as the number of particles and the density of collision bodies). Simultaneously, rendering and physics tasks share the GPU memory space, and only the instruction pointer is reset when switching, without the need for data migration, and the switching delay is <0.1ms. In addition, if an urgent task is detected in the physical slot (such as particle generation for explosion special effects), the current rendering slot is immediately preempted (preemptive execution).

[0062] To further balance the computational efficiency between 3D model rendering and physics processing, in the dynamic computing power allocation module, when GPU utilization is ≥95%, a hierarchical degradation strategy is initiated. Specifically, the following are included: Level 1 degradation: turning off ray tracing reflections and global illumination; Level 2 degradation: switching the physics simulation accuracy to low-resolution proxy mode; Level 3 degradation: freezing the physics calculations of non-interactive background models. This hierarchical degradation strategy gradually reduces the rendering and physics calculation load (turning off high-energy-consuming special effects → switching to proxy mode → freezing non-core calculations), dynamically releasing resources when the GPU is overloaded, prioritizing the real-time and stability of core interactive tasks (such as character movements and collision feedback), avoiding system freezes or crashes, and achieving a controllable sacrifice in rendering quality and physics accuracy. This ensures that a minimum frame rate of 60fps and an interactive delay of less than 10ms can still be maintained under extreme loads, balancing efficiency and user experience.

[0063] The cross-scene adaptation module is used to call upon a library of preset physical parameter templates to adapt to different AIGC-generated scenes. It also seamlessly stitches the 3D models generated by the AIGC generation module into animation sequences for different scenes. The cross-scene adaptation module also implements physical parameter migration through keyword-driven template switching. The specific steps are as follows: a. Parse the scene keywords input into the AIGC generation module; b. Match the physical parameter sets in the preset template library; c. Complete parameter loading within 100ms and seamlessly integrate them with the current animation sequence.

[0064] The specific implementation process is as follows: first, a text instruction "Generate an astronaut tumbling in a zero-gravity environment" is input into the AIGC generation module. The AIGC generation module generates a three-dimensional model of the astronaut (with 100,000 faces) and a tumbling animation sequence through the diffusion model.

[0065] At the same time, the physical constraint training model is optimized in real time, generating the following results:

[0066] Center of mass projection error loss: When constraining tumbling, the trajectory of the astronaut's center of mass complies with the law of conservation of angular momentum, eliminating non-physical rotational acceleration;

[0067] Collision penetration loss: The SDF distance field is used to detect penetration events between astronauts and virtual bulkheads, and the astronaut model implements a 1.1x collision volume expansion;

[0068] Joint torque loss: The inverse kinematics solver is used to limit the rotation angles of the limb joints (elbow joint ≤ 140°, knee joint ≤ 120°) to prevent biomechanical distortion.

[0069] Continuing with dynamic computing power allocation and physics verification, the physics time slot was set to 10ms (the FPGA computes the rigid-body collision between the astronaut and the cabin wall in zero gravity (response latency <1ms)), and the rendering time slot was set to 6ms (the GPU performs ray tracing (reflections, shadows) and particle effects (breathing mist)). The astronaut's hands (high-frequency interactive area) retained a full-precision mesh (100,000 faces) with a physics calculation frequency of 120Hz. The mesh precision of the background console (low-frequency non-interactive area) was reduced to 20% (20,000 faces), freeing up 30% of GPU resources.

[0070] Then, the cross-scene adaptation module extracts the "zero gravity" keyword and loads parameters from the template library (gravity coefficient = 0m / s 2 , air resistance = 0.01N·s 2 / m 2 When the physics engine detects penetration between the astronaut and the cabin wall, it triggers the AIGC generation module to complete local corrections before the next frame (leg bone rotation angle -3°, collision volume expansion 1.05 times). The correction data is directly transmitted to the rendering engine via CUDA Stream, and the screen is smooth (delay <5ms).

[0071] A fine mesh (high-resolution collision volume) with 1,000 faces per cubic decimeter is generated for the astronaut's hand contact area to calculate precise grip force feedback. A spherical proxy (100 faces) is used for the cabin wall, reducing collision detection time by 80%. When GPU utilization reaches 95%, a tiered downgrade strategy is implemented: Level 1: ray tracing reflections are disabled, freeing up 15% of resources; Level 2: The cabin wall is switched to a bounding box proxy volume, reducing the physics calculation load by 40%; Level 3: The physics simulation of the background starry sky is frozen to ensure a core interactive frame rate of ≥60fps.

[0072] The resulting animation sequence, depicting the astronaut's tumble, complies with the law of conservation of angular momentum, with a center-of-mass trajectory error of less than 0.1 cm and a penetration event rate reduced from 12% to 0.5% compared to traditional approaches. The cross-scenario adaptation module also supports parameter template switching, enabling seamless transitions from "zero gravity" to "underwater fluid" scenarios.

[0073] Table 1-Test index comparison table

[0074]

[0075] As shown in Table 1, this solution significantly outperforms traditional solutions in terms of generation efficiency, physical rationality, real-time performance, and resource utilization through physical constraint pre-training, heterogeneous computing acceleration, and dynamic LOD optimization. It achieves higher stability (frame rate fluctuation <±2fps, penetration rate <0.5%) in particularly complex scenarios (such as explosion effects and zero-gravity interactions).

[0076] Example 2:

[0077] The difference from the above embodiment is that, Figure 2 As shown, a 3D animation effect simulation method based on AIGC, the specific steps are as follows:

[0078] Step 1: AIGC generation and physical constraint training. Based on the input instructions, the AIGC generation module generates a three-dimensional model and animation sequence. During the generation process, the output results are optimized in real time through the physical constraint training model.

[0079] In step 2, the physics engine verification module detects the physical rationality of the animation sequence frame by frame; based on the complexity of the scene, tasks are dynamically allocated through the collaborative architecture of heterogeneous computing units.

[0080] Step 3: Cross-scene parameter migration and local correction. Analyze the scene keywords in the AIGC input and match the physical parameter set from the preset template library; and perform local parameter correction in areas where physical conflicts are detected.

[0081] Step 4: Multi-resolution physical mesh and dynamic LOD optimization. Generate a multi-resolution physical network for the 3D model, and generate high-resolution collision bodies or low-resolution proxy bodies based on the physical interaction frequency and scene requirements. In the animation sequence, the model parts where the frequency of physical interaction between the 3D model and user operations or the environment exceeds the preset threshold will retain the full-precision mesh and increase the physical calculation frequency to 120Hz. The mesh accuracy of the 3D model far away from the collision area will be reduced to 10%-30% of the original 3D model.

[0082] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A 3D animation effect simulation device based on AIGC, characterized in that: include: AIGC generation module, used to generate 3D models and animation sequences based on input instructions; The physics engine verification module is used to verify the physical rationality of 3D models in real time based on animation sequences, including collision detection, rigid body dynamics, and center of mass stability calculations; The computing power dynamic allocation module is used to dynamically allocate GPU resources to rendering or physical simulation tasks according to scene requirements; The cross-scene adaptation module is used to call the preset physical parameter template library to adapt to different types of scenes generated by AIGC, and seamlessly stitch the three-dimensional model generated by the AIGC generation module with the animation sequences of different scenes.

2. The 3D animation effect simulation device based on AIGC according to claim 1, characterized in that: During the training phase of the AIGC generation module, the output model of the AIGC generation module adopts a physical constraint training model; The physical constraint training model includes: center of mass projection error loss unit, collision volume penetration loss unit and joint torque rationality loss unit; The centroid projection error loss unit is used to constrain the three-dimensional model's motion to conform to static equilibrium; The collision volume penetration loss unit is used to block non-physical overlap between 3D models; The joint torque rationality loss unit is used to constrain the limb motion of the three-dimensional model to conform to biomechanics.

3. The AIGC-based 3D animation effect simulation device according to claim 1, characterized in that: The dynamic computing power allocation module divides the GPU computing cycle through hardware-level time slice rotation, and divides the computing unit into physically isolated rendering partitions and physical partitions based on the GPU hardware scheduler. Each partition exclusively uses the SM cluster and L2 cache resources, and creates an independent task queue through CUDA Stream.

4. The 3D animation effect simulation device based on AIGC according to claim 1, characterized in that: The AIGC generation module works in conjunction with the physics engine verification module to perform dynamic level of detail adjustments; Dynamic Level of Detail adjustments include: Reduce the mesh accuracy of the 3D model away from the collision area to 10%-30% of the original 3D model; For model parts in animation sequences where the frequency of physical interaction between the 3D model and user operations or the environment exceeds a preset threshold, the full-precision mesh is retained and the physical calculation frequency is increased to 120Hz.

5. The 3D animation effect simulation device based on AIGC according to claim 4, characterized in that: The cross-scenario adaptation module is also used to implement physical parameter migration through keyword-driven template switching. The specific steps are as follows: a. Parse the scene keywords of the input text in the AIGC generation module; b. Match the physical parameter set in the preset template library; c completes parameter loading within 100ms and seamlessly integrates with the current animation sequence.

6. The AIGC-based 3D animation effect simulation device according to claim 5, characterized in that: The physics engine module is also used to complete the following operations within a single frame period: detecting physical conflicts in the 3D model; triggering the AIGC generation module to perform local parameter corrections on the conflicting area; Correction methods include: adjustment of limb rotation angle; Compensation for collision volume expansion.

7. The 3D animation effect simulation device based on AIGC according to claim 6, characterized in that: The dynamic computing power allocation module adopts a collaborative architecture of heterogeneous computing units and automatically switches computing units according to the rendering / physics task type.

8. The AIGC-based 3D animation effect simulation device according to claim 7, characterized in that: The physics engine verification module is also used to generate multi-resolution physical meshes for 3D models; Multi-resolution physical grids include: high-resolution collision volume, grid density ≥ 1000 faces / cubic decimeter; low-resolution proxy volume, grid density ≤ 100 faces / cubic decimeter; The low-resolution proxy body and the high-resolution mesh are synchronized in motion using a deformation mapping algorithm.

9. The 3D animation effect simulation device based on AIGC according to claim 8, characterized in that: In the dynamic computing power allocation module, when the GPU utilization rate is ≥95%, the hierarchical degradation strategy is activated; Tiered downgrade strategies include: Level 1 downgrade: turn off ray tracing reflections and global illumination; Second level degradation: switches the physics simulation accuracy to low-resolution proxy mode; Level 3 degradation: Freeze the physics calculations of the non-interactive background model.

10. A 3D animation effect simulation method based on AIGC, according to the 3D animation effect simulation device based on AIGC according to any one of claims 1 to 9, characterized in that: The specific steps are as follows: Step 1: AIGC generation and physical constraint training: Based on the input instructions, the AIGC generation module generates a 3D model and animation sequence. During the generation process, the physical constraint training model is used to optimize the output results in real time. Step 2: The physics engine verification module checks the physical rationality of the animation sequence frame by frame; Dynamically allocate tasks based on scenario complexity through a collaborative architecture of heterogeneous computing units; Step 3: Cross-scenario parameter migration and local correction: parse the scene keywords in the AIGC input, match the physical parameter set from the preset template library, and perform local parameter correction in areas where physical conflicts are detected. Step 4: Multi-resolution physical mesh and dynamic LOD optimization. Generate a multi-resolution physical network for the 3D model, and generate high-resolution collision bodies or low-resolution proxy bodies based on the physical interaction frequency and scene requirements. In the animation sequence, the model parts where the frequency of physical interaction between the 3D model and user operations or the environment exceeds the preset threshold will retain the full-precision mesh and increase the physical calculation frequency to 120Hz. The mesh accuracy of the 3D model far away from the collision area will be reduced to 10%-30% of the original 3D model.

Images