AEmotionStudio/audiox-models

AudioX Models (Safetensors)

.safetensors conversions of AudioX-MAF model checkpoints for use with ComfyUI-FFMPEGA.

AudioX is a unified anything-to-audio model from ICLR 2026 that supports text-to-audio, text-to-music, video-to-audio, and audio inpainting.

Models

File	Description	Size
model.safetensors	AudioX-MAF DiT model (full precision)	5.19 GB
synchformer_state_dict.safetensors	Synchformer temporal encoder (shared with MMAudio)	475 MB
config.json	Model architecture configuration	3.3 KB

Sources

All models were downloaded from their original sources and converted by us:

• AudioX-MAF: HKUSTAudio/AudioX-MAF
• Synchformer: Shared with MMAudio

Usage

These models are automatically downloaded by the generate_music and audio_inpaint skills in ComfyUI-FFMPEGA. No manual setup needed.

Manual installation:

ComfyUI/models/audiox/
├── model.safetensors
├── synchformer_state_dict.safetensors
└── config.json

Note: The synchformer_state_dict.safetensors is shared with MMAudio. If you already have it in ComfyUI/models/mmaudio/, AudioX will reuse it automatically — no duplicate download needed.

Capabilities

Skill	Description
generate_music	Text-to-music and video-to-music generation
audio_inpaint	Fill gaps, extend, or regenerate sections of audio

License

⚠️ CC-BY-NC 4.0 — AudioX model weights are licensed under Creative Commons Attribution-NonCommercial 4.0. Commercial use of the models is restricted. The code that loads/runs them is GPL-3.0.

Paper

AudioX: Diffusion Transformer for Anything-to-Audio Generation (ICLR 2026) arXiv:2503.10522