These are simply a merged in version on top of Flux 2 Klein 9b Base or Distilled with a little extra training for those that can't use lokrs for whatever reason. 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 (parts of which shouldn't really matter because Flux 2 Klein is under a noncommercial 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!
Description
FAQ
Comments (23)
What are the parameters to get quality images? I see that you used a few samplers, but nothing about schedulers, steps, or CFG.
I just kept it default-ish for the samples. cfg of 5, steps of 50 (though I'm sure you could go lower), Flux2Scheduler node for the scheduler.
It was trained up to 1.5mp as far as resolution goes but it can go higher.
I'm freaking lovin your lora for klein 9b of this. Is there any difference by using this as checkpoint rather than the default flux checkpoint+lora?
Not sure if he updated the description but it says
"This is simply a merged in version on top of Flux 2 Klein 9b Base with a little extra training for those that can't use lokrs for whatever reason"
Not really, no.
Is there a full model that loras can be trained on top of?
Need fp8 version for 32 Gb RAM PCs.
Actually using this checkpoint + only a lora character (plus Klein 4B/9B Base to Turbo Lora with 1.0 weight and 4 steps) gives outstandingly better results and notieceably less body horror than using the SNOFS lora + base model + char lora. I highly recommend this setup
Hi, how did you train your character Lora ?
@jeremyhola ai toolkit default flux klein 9b settings
Is there a diference between using the Klein 9B distill and the Klein 9B Base + Base to Turbo Lora?
Nice work. Thanks to your openness I was able to finetune (full) my model successfully and when it is done I will upload. If you want to collaborate on datasets and techniques let me know.
this might be great base to train Flux 2 LORA on that needs NSFW knowledge
Perfect combo with DR34MSC4PE lora 0.69 weight
What is a lokr? Do you mean lora?
No, I meant lokr.
"Low-Rank Kronecker Product (LoKr), is a LoRA-variant method that approximates the large weight matrix with two low-rank matrices and combines them with the Kronecker product. LoKr also provides an optional third low-rank matrix to provide better control during fine-tuning."
Any chance you have the distilled version yet?
Chiming in to say, I'm also very eager for the distilled version
Sorry, not yet - my upload speed isn't great and I've had a few failures to upload. I'll try again tonight.
It's up
@Ashen3 can i add you for direct convo, discord or telegram whatever? thx boss u do a rly good job
@Ashen3 I may be wrong but I think most people (including myself) were expecting an fp8 distilled model.
you did an amazing job, thank you very much and don't stop :)