Simple Fix: DONT PAY. That's all on you.

don’t be such a simp

On weight 0.8 work perfectly even with Scnell on 5 steps

Did you by chance add "Horny" to your prompt?

flux doesn't have negative prompts

I believe that's because originally FLUX is hardcoded for very soft nsfw only (if any) and tries to prevent any restriction violations. Some nsfw Lora did help, but not nearly as much as good SDXL models.

We need to re-train FLUX without the harcoded tags, or we need some inside original dev team to distribute ( under the table ) .... the unlocked version! XD

Captioning seems to be very important for flexible Flux training, especially when introducing new concepts like specific NSFW concepts. I have been using the Joy caption tool based on Llama with some uncensored Llama models and contextual prompts for each subdirectory. When training, Flux picks up on artifacts and textures first, so including any blurry or hyper-realistic images will bias the training in that direction.

@AIDigitalMediaAgency No problem!

@AIDigitalMediaAgency will check it again.

use Q8 works great