📘 I’m preparing documentation on creating models using unique methods.
🎁 Active users of my models will receive it for free.
Follow and share your creations.
BRAVO – REALLUX
For beginners:
Supports simple prompts and works even without a negative prompt.
For experienced users:
Use film and optical terms in your prompt for better control.
For professional use:
Unlike many other models, this one correctly generates portraits from 2–5 reference photos.
Link: Result & Settings
Files: https://huggingface.co/h94/IP-Adapter-FaceID/tree/main
For the best likeness, it is recommended to specify age (young, middle-aged, elderly) and body type (slim, chubby, plus-size, overweight).
For these purposes, I use AnyLoRA; the result is more similar to the originals.
Recommended settings:
Schedule type: Karras
Sampling method: DPM++ 2M
CFG Scale: 4–7
Sampling steps: 20 (for simple prompts, increasing steps may worsen the result)
Description
FAQ
Comments (2)
This model is incredible.
It doesn't require any VAE.
It works very well with minimal instructions, even without negative prompts.
Great quality. I don't use any hand enhancements, and the model still always tries to create four knuckles (not counting the thumb).
Seriously, it's very well done. I'm giving you 10 yellow Buzz. But if yellow Buzz weren't so scarce as they used to be, I would have given you 100.
═══
Euler a
clip skip: 1
steps: 40
512 x 824
DFG: 4
Thank you, esteemed colleague!
"Schedule type: Beta" tends to generate images with a more creative approach. I recommend you try it with your current settings.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.