https://github.com/asutermo/ComfyUI-Flux-TryOff
https://github.com/asutermo/ComfyUI-Flux-TryOff
Last synced: 3 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/asutermo/ComfyUI-Flux-TryOff
- Owner: asutermo
- License: mit
- Created: 2025-01-21T18:36:28.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2025-01-21T23:12:52.000Z (3 months ago)
- Last Synced: 2025-01-22T00:20:37.532Z (3 months ago)
- Language: Python
- Size: 24.4 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
Awesome Lists containing this project
- awesome-comfyui - **ComfyUI-Flux-TryOff**
README
# ComfyUI-Flux-TryOff
Original Source: [catvton-flux](https://github.com/nftblackmagic/catvton-flux). I implemented their try-off inference code as ComfyUI nodes
There's a sample workflow in [Workflow](https://github.com/asutermo/ComfyUI-Flux-TryOff/tree/main/workflow) that uses SegFormer to generate the mask for you. I highly recommend this approach. Alternatively you can provide your own!
Please note, that this was tested with a 4080, and it's quite slow. You'll want a 4090 or better for performant execution as of right now.
This uses diffusers>=0.32.0
After heavy experimenting with Try-on, it's nice to have a [Try-Off, xiaozaa/cat-tryoff-flux](https://huggingface.co/xiaozaa/cat-tryoff-flux) model to work with.
The cat-try-off-flux model will download automatically. The Flux.1-dev model requires some effort.
1. Go to huggingface
2. Go to your settings and generate a 'write' token
3. Go to https://huggingface.co/black-forest-labs/FLUX.1-dev and accept the terms
4. Open a prompt, go to your ComfyUI installation and do the following
Windows
```bat
SET HF_TOKEN=
SET HUGGING_FACE_HUB_TOKEN=
```
Linux
```sh
EXPORT HF_TOKEN=
EXPORT HUGGING_FACE_HUB_TOKEN=
```
Finally, download FLUX.1
```sh
cd ./models/checkpoints
git lfs install
git clone https://huggingface.co/black-forest-labs/FLUX.1-dev
```
## TODO
- Optimize, optimize, optimize.
- Allow additional models
- Formatting/consistency
- Precision