---
license: other
license_name: sam-license
license_link: LICENSE
tags:
- comfyui
- audio-separation
- sam-audio
- audio
- flow-matching
pipeline_tag: audio-to-audio
---

# ComfyUI-SamAudio

Converted standard `BF16` `.safetensors` checkpoints for **SAM-Audio** (Segment Anything Model for Audio) by Meta AI, optimized for native use in **ComfyUI**.

## Available Checkpoints

| Checkpoint | Format | Size | Description |
| :--- | :---: | :---: | :--- |
| **`sam_audio_base.safetensors`** | BF16 | 1.73 GB | 12-layer DiT, 1536 dim. Fast daily driver. |
| **`sam_audio_large.safetensors`** | BF16 | 6.38 GB | 22-layer DiT, 2816 dim. Maximum separation fidelity. |

## Installation & Placement in ComfyUI

Place the `.safetensors` model(s) into:
```
ComfyUI/models/sam-audio/
```

Quick download via CLI:
```bash
hf download Aero-Ex/ComfyUI-SamAudio --local-dir ComfyUI/models/sam-audio
```

## Required Text Encoder

For text prompting, place `t5-base.safetensors` into `ComfyUI/models/text_encoders/`:
- Download: [Comfy-Org/t5-base](https://huggingface.co/Comfy-Org/t5-base/blob/main/t5-base.safetensors)

## Custom Node

Run these models natively using the dependency-free custom node:
- **`ComfyUI-SAM-Audio-Native`**

### Supported Modalities
- **Text-Prompted Separation:** Isolate vocals, speech, instruments, sound effects using natural language.
- **Temporal Span Separation:** Isolate sounds using `+` (present) or `-` (absent) time intervals.
- **Video-Guided Separation:** Isolate sounds tied to visual objects using video frames and tracking masks (e.g. from SAM 3).

## License

This model is subject to Meta's SAM License. See [LICENSE](LICENSE) for details.