Improved Female Nudity Lora
Flux Klein 9B
Version 3.0 has no trigger word
My goal here was to create photorealistic nudes, not just idealized bodies. Things like detailed skin pores, various breast types and body shapes, I think v2.5 is closer to that overall goal. Lora was trained on natural flux language, I typically just use the 8 step distilled model
Sampler: Try euler or euler_a, dpmpp_sde is another good one
Model: all my images use flux-2-klein-9b-BF16.gguf (distilled version) with cfg 1.0, if you have trouble just drag one of my images into comfyUI and run the workflow, I use a black 768x1280 image as my latent
version 2.5 was trained on 340 images, with detailed prompts focused on different poses, lighting conditions, age groups, ethnicities etc
Description
version 2.5 retrained to produce higher quality outputs
FAQ
Comments (14)
Nice!
It would be nice to see such types of LORA for different body types: fit, muscular, skinny etc.
give v3 a try?
Does this also work on edit?
text-to-image and image edit are the same model so yes
@sexgod1979 it does absolutely not work in edit. just tried. the qwen version works but it always gives belly piercings.
@kronos1959777 hit me on chat and ill assist you
@sexgod1979 Was this lora focused on T2V? The face changes quite a lot when using your prompts and workflow. Only when I trimmed back the prompts to first few lines do I get closer to face consistency.
@rumbleskin I dont run into any issues with img 2 img in klein 9b using this lora keeping faces locked, it works well for me similar to qwen. However to answer your question yes it was trained without control images, purely for T2I (text to image). Its meant to generate images purely from text. civitai doesnt like edit loras, so I wont bother likely
@rumbleskin and if you want to use this lora for edit you need to keep the prompts short and sweet, things like 'exact face from image 1, blah blah blah on a beach' thats it. Dont word soup it, thats just for text to image stuff
How do you generate your prompts? some are quite detailed.
sometimes use a LLM (AI) to generate prompts so I get good training, other times I Just write them by hand as ive been writing flux captions for for a while. Sometimes ill give an existing photo to AI/LLM and then ask it to caption it for me and use that to generate a new image
you can try with grok, it won't snitch
FOR 4B