Seeing that FLUX seems to struggle with depicting armpit hair, I decided to create my first ever concept LoRA. The model was trained exclusively on photos of hairy armpits of women of various ethnicities and ages.
Recommendations
Use LoRA weight 0.4 to 0.7 as a starting point.
Increasing or decreasing the weight increases or decreases the pronouncedness of armpit hair.
Increasing the weight increases the influences other features of the image.
Supports long, medium, and short armpit hair.
Supports all armpit hair colors, e.g. black, brown, blonde, auburn, turquoise, pink.
Can be used for inpainting!
Works with other LoRAs (e.g. people, styles).
Works with FLUX.1-dev (including fp8 and other quantized versions).
CFG 2 to 4 seems to be fine.
Description
v1.0 was trained on 84 photos of various women with various amounts of armpit hair in various SDXL compatible resolutions.
FAQ
Comments (4)
Doesn't seem to work for me. Can you give some prompt example? I've put it to maximum 2.0 and yet no armpits hair
It should be triggered by "(short/medium/long black/brown/blonde) hairy armpits". Because the dataset was somewhat limited, try to include "holding one arm behind her head" or "showing an armpit". This will be amended in a future v2.0. Please note, that while I tested the LoRA with fp16, fp8,, and Q_6, it might not work when using --fast (in ComfyUI) or a TorchCompileModel node.
@a__a im still not getting any results. What Sampler did you use?
@kamik4ze422 That's odd. I'm using the XlabsSampler node in ComfyUI, but it should work with any KSampler and scheduler. If you'd like to share your workflow with me, maybe I'll find what's causing this.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.