Flux Unchained by SCG

Total training was about 5k images spread across SFW cinematic stills, art photography, LaION art-pop, about 1500ish explicit and artful nudes about 80% photography, 20% AI/illustrative nudes. The model responds well to prompts just like base-flux does. This is a WIP and only a V1, I will be tuning this model more as I identify weaknesses in the output and methods to improve the quality. This model was built on top of the flux.1_dev_8x8_e4m3fn-marduk191 version, so fp8 quality, with FP8 CLIP-L and T5xxl Encoders to get the model size down just a bit and hopefully make it usable for folks with 16GB of VRAM or less (good luck!)

(the only difference between this model and the t5_16x8 model is the text encoder, there is no need to keep both around)

Known Issues

• No male genitalia yet. have a lora for it, but it messes up lady bits currently, needs redesigning.

• this model is horny, and reminds me a bit of 1.5 now, in that it will sometimes bias towards nudity unprompted. Will work on this with future versions, but for the time being, it's a good idea to prompt clothing if you don't want nudity.

• Super fat guys sometimes get boobs. It's actually realistic I guess, but still looks a bit weird.

• I've noticed some odd striping on the sides of some images in some aspect ratios. This is something the base flux.d model does as well, though it seems the pattern has changed a little with my tune. Will keep an eye on this, may be related to how I'm building the model, or bucketing my training inputs.

• As this is trained on much of the same photo and cinematic library that NightVisionXL is trained on, there is going to be a bias towards closer portraits.

• Women's naked hips are sometimes displaying a wierd ribbing pattern. This is due to the model still really really wanting to put some kind of underwear or bottoms. This should clear up as I layer more trainings in, but it's a bit of an annoyance currently.