PuLID-Flux ComfyUI implementation (Alpha version)
- V0.1.0: Working node with weight, start_at, end_at support (attn_mask not working)
This project was heavily inspired by cubiq/PuLID_ComfyUI. It is just a prototype that uses some convenient model hacks
for the encoder section. I wanted to test the model’s quality before reimplementing it in a more formal manner. For better results I recommend the 16bit
or 8bit GGUF
model version of Flux1-dev (the 8e5m2 returns blurry backgrounds).
In the examples
directory you'll find some basic workflows.
For GGUF models you will need to install ComfyUI-GGUF
- Install this repo into
ComfyUI/custom_nodes
git clone https://github.com/balazik/ComfyUI-PuLID-Flux.git
- You need one of the mentioned
Flux.1-dev
models. Download the model intoComfyUI/models/unet
, clip and encoder intoComfyUI/models/clip
, VAE intoComfyUI/models/vae
. - PuLID Flux pre-trained model goes in
ComfyUI/models/pulid/
. - The EVA CLIP is EVA02-CLIP-L-14-336, should be downloaded automatically (will be located in the huggingface directory).
facexlib
dependency needs to be installed, the models are downloaded at first use.- Finally you need InsightFace with AntelopeV2, the unzipped models should be placed in
ComfyUI/models/insightface/models/antelopev2
.
- ApplyPulidFlux doesn't work on HW with CUDA compute < v8.0, (when Flux FP8 it needs bfloat16).
- When the ApplyPulidFlux node is disconnected after first run, the Flux model is still influenced by the node.
- ApplyPulidFlux attn_mask is not working (in progress).
ComfyUI/ComfyUI - A powerful and modular stable diffusion GUI.
PuLID for Flux - tuning-free ID customization solution for FLUX.1-dev
cubiq PuLID_ComfyUI - PuLID ComfyUI native implementation (Thanks for the awesome work what you do Matteo 😉 ).