---
license: mit
tags:
- whisper
- speech-recognition
- transcription
- comfyui
- ffmpega
---

# OpenAI Whisper Models

Mirror of [OpenAI Whisper](https://github.com/openai/whisper) model weights in **safetensors** format, hosted by [AEmotionStudio](https://github.com/AEmotionStudio) for use with [ComfyUI-FFMPEGA](https://github.com/AEmotionStudio/ComfyUI-FFMPEGA).

## Available Models

| File | Size | Parameters |
|:-----|:-----|:-----------|
| `tiny.safetensors` | ~72 MB | 39 M |
| `base.safetensors` | ~139 MB | 74 M |
| `small.safetensors` | ~461 MB | 244 M |
| `medium.safetensors` | ~1.5 GB | 769 M |
| `large-v3.safetensors` | ~2.9 GB | 1550 M |

**Original source:** [OpenAI Whisper](https://github.com/openai/whisper) (Azure CDN)

## Format

Models are stored in `.safetensors` format for security. The `dims` configuration metadata is preserved in each file's safetensors header. ComfyUI-FFMPEGA automatically handles the conversion to `.pt` format locally when downloading from this mirror.

## Usage

These models are automatically downloaded by ComfyUI-FFMPEGA when audio **transcription** or **subtitle generation** is used. No manual setup is required if `allow_model_downloads` is enabled on the FFMPEG Agent node.

### Manual Installation

1. Download the desired `.safetensors` file from this repo
2. Convert to `.pt` or place in `ComfyUI/models/whisper/`
3. FFMPEGA will detect the local file and skip downloading

## Why This Mirror?

This is a first-party mirror maintained by AEmotionStudio to ensure supply chain resilience. If OpenAI's CDN becomes unavailable, FFMPEGA will still be able to download models from this mirror.

## License

OpenAI Whisper is released under the MIT license.