Flux.2 Klein - 3 Contemporary Pulid v2.1a

Hi! The other versions went MIA in the Civitai database, so here is the fresh, updated release!

With the PuLID system, you can generate a new Text-to-Image (T2I) render using the exact head and face of your choice with surgical precision.

As always, you will have to experiment a little with your prompt or add some LoRAs if you need them (though this workflow doesn't strictly require any to function). Feel free to use your favorite version of Flux.2 Klein to match your creative ideas.

• Settings: CFG 1, Steps 8-10 (4 steps for a quick preview, but 8-10 is the sweet spot), Denoise 1.0... it's Klein! PuLID Strenght: i've used 1.4 for this image but i's better you start from 1.0 and see or you may have very hugly generations!!!

• Multi-Character: Inside, you will find 3 PuLID setups so you can add up to three different characters in a single generation.

Important Note on Face Images & Memory (VRAM):
You will obviously need good quality reference photos. If the head is too small, the model can't work miracles. Because I prefer prepping my face images beforehand, I wanted to make things easier for those who don't: I have added a Remove Background node and a Crop Face node for each of the three head inputs.

On your first run, ComfyUI will automatically download the required background removal model. I have explicitly chosen BEN2 (BiRefNet v2) because it perfectly balances edge quality with memory usage. Do NOT try to switch it to RMBG-2.0—running three instances of RMBG-2.0 in a batch will send your system straight to an Out of Memory (OOM) crash in a few seconds!

If I forgot something, let me know in the comments.

In the next few days, I’ll share a "new" version of an editing workflow. It features optimized prompts and LoRAs designed to dramatically improve small, low-quality, or blurry photos and upscale them in a single step, making them perfect to use as references for this workflow or your other projects.

Have fun! 😊