These are simply a merged in version of SNOFS on top of Krea 2 Raw or Flux 2 Klein 9b Base or Distilled can't use lokrs for whatever reason, or want to train on top of them instead of the base model (so characters or styles come out well). I suppose you could also use them to generate on Civitai - if you want to run the Krea version as turbo on Civitai use one of the turbo extracted loras with it (and then CFG 1, 8 steps with Euler/Simple).
For full details on how best to use this model please check out the lokr page: https://civarchive.com/models/1972981/sex-nudes-other-fun-stuff-snofs
Both the lokr and this trained model are released under this license:
https://huggingface.co/Ashen3/SNOFS
Want to support my work or help fund the training of this dataset on other models? Join the Patreon in my profile, and if you do - thank you!
SNOFS Krea 1.3D bf16 and int8 convrot are both available as variants.
Description
FAQ
Comments (22)
fantastic checkpoint, thanks
so we use it as diffusion model or lora?
I don't think you will be using a 16GB model as a LoRA :D The description says this is essentially his LoRA merged with the base model - so this is basically now a hornier Klein model.
@altoiddealer AHAAHA, alr so i use it in replace for klein 9b or klein9b true, amazing work !
This can't read at Forge Neo.
Wont detect in Comfy
Apparently ComfyUI made a change to their model save node where it now saves in ComfyUI's internal format. I'd fix these files but I should have the new model out soon, so the next version will be fixed. Sorry about that.
is anyone getting this error:
ERROR: Could not detect model type of: snofsSexNudesAndOtherFunStuff_distilledV1.safetensors
yes
if the creator saved the checkpoint in comfyui it is normal because the keys are wrong then
I tried using a different clip from Qwen 3 8B, switching to it fixed it for me.
@MrEsado What were you using before?
Same error here.
Appreciate the models - though it would be great to get an fp8 distilled version.
Good job though, this is really nice. :D
yes waiting for the fp8 version as well
Your wishes are my command.
Does this apply to edit mode?
Make one for new black forest labs 9B KV version please 🙏 its new hot thing
Oh thats new, need to test KV version, thanks :)
Is it possible to train a character lora on this? Aitoolkit had issues with this checkpoint.
Yes, it's possible to use this as your model in AI Toolkit (and other LoRA training software). You need to fix the keys first though. The following Python script will fix the keys by removing the prefix "model.diffusion_model."
from safetensors.torch import load_file, save_file
src = "snofsSexNudesAndOtherFun_v10.safetensors"
dst = "snofsSexNudesAndOtherFun_v10_FIXED.safetensors"
state_dict = load_file(src)
new_state_dict = {}
for k, v in state_dict.items():
new_k = k.replace("model.diffusion_model.", "")
new_state_dict[new_k] = v
save_file(new_state_dict, dst)
I should note that this worked for version 1.0 as that's the full model. I have not tried it with the distilled version and I am doubtful that will work well.