THA3 TensorRT
THA3 animates an anime character from a single portrait. Given a 512 × 512 transparent RGBA image and a 45-element pose vector, it produces a new frame with the requested facial expression and head or body pose.
This release provides tha3-tensorrt-fp16.onnx, an ONNX export of the pretrained separable_float model adapted for TensorRT 10.4 with mixed FP16 precision.
License and Attribution
Original work by Pramook Khungurn (2022), Talking Head Anime from a Single Image 3: Now the Body Too. Project page · Source repository
The model weights are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). This release converts the original pretrained model to mixed FP16 ONNX and adapts the graph for TensorRT.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support