部署美团 LongCat-Video-Avatar (InfiniteTalk v1.5),实现单图+音频 → 数字人唱歌视频。
�ash git clone https://github.com/stellanoxUN/longcat-setup.git cd longcat-setup bash setup.sh
�ash cd LongCat-Video torchrun --nproc_per_node=1 run_demo_avatar_single_audio_to_video.py \ --input_json config.json \ --checkpoint_dir ../weights/LongCat-Video-Avatar-1.5 \ --model_type avatar-v1.5 \ --resolution 480p \ --num_segments 38 \ --output_dir ./results
json { "prompt": "A female singer in a recording studio singing emotionally", "cond_image": "path/to/image.png", "cond_audio": { "person1": "path/to/audio.wav" } }
- GPU: 16GB+ VRAM (推荐 RTX 4090)
- 磁盘: 80GB+
- CUDA: 12.4
- Python: 3.10+
- 国内机器会自动使用 hf-mirror.com 镜像下载