Solutions for unnatural movements
ChatAnyone uses Hierarchical Motion Diffusion Modeling technology to specifically optimize the naturalness of digital human movements. Below are the specific solutions:
- Import high-quality photos: Ensure that portraits are clear, front-face oriented, and avoid occlusions. Official examples show that it's easier to create natural head turns and expressions in front-face portraits.
- Optimized audio input: Use audio files with clear pronunciation and moderate speech rate. The system generates lip synchronization and micro-expressions according to the audio tempo
- Motion range control: In the current demonstration, the system presets a reasonable range of motion amplitude (e.g., 15°-30° head rotation) to avoid exaggerated distortion.
- Hardware Matching: High-performance GPUs (e.g. 4090) are required to ensure computing power for smooth output at 30FPS
Alternative: If open source can try to adjust the model parameters, through the training data to optimize the specific style of action. At this stage, we can observe the effect of different speech inputs through the official demonstration video.
This answer comes from the articleChatAnyone: a tool for generating half-body digital human portrait videos from photosThe
May not be reproduced without permission:AI productivity tools " How to solve the problem of unnatural movements when generating digital human videos from photos?
