henrybit's picture
Build HunyuanVideo-Avatar ZeroGPU evaluation app
b653e58 verified
|
Raw
History Blame Contribute Delete
1.39 kB

A newer version of the Gradio SDK is available: 6.26.0

Upgrade
metadata
title: HunyuanVideo Avatar Test
emoji: 🎭
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pytorch_version: 2.8.0
python_version: '3.10'
pinned: false
license: other
suggested_hardware: zero-a10g
short_description: Test cartoon avatars with HunyuanVideo-Avatar

HunyuanVideo-Avatar ZeroGPU Evaluation Space

Evaluation app for the official Tencent HunyuanVideo-Avatar single-GPU inference path.

Evaluation profile

  • One avatar image + one speech audio file + an English scene prompt
  • Official FP8 checkpoint
  • 704px inference profile
  • 49 / 73 / 97 / 129 frame test lengths at 25fps
  • 50 inference steps with DeepCache
  • CPU group offload
  • @spaces.GPU(size="xlarge") only on the generation function

The upstream repository is approximately 80.8GB. This Space downloads only the required FP8 and inference assets (approximately 45GB) and excludes the full precision transformer plus duplicate TensorFlow, Flax, and PyTorch model files.

The first request can take a long time because the weights are downloaded lazily. HunyuanVideo-Avatar is a very large model; the 49-frame mode is recommended for initial compatibility and visual-quality testing.

Use only images and audio you have permission to process. Clearly disclose AI generated or altered media when publishing the result.