A nude/NSFW capable model that focuses on images with a female primary subject.
It is heavily biased towards nude images.
Recommended Settings
Sampler is dpmpp_2m (DPM++ 2M) and the scheduler is beta
No VAE or CLIP is baked in. Use separate sources for those.
Each sample/preview image contains the used workflow. Here's a quick article with simpler more beginner-friendly workflow. This is a recommended starting point.
FP8 Download
For some reason, when you upload multiple formats (BF16 + FP8) on one model, which they now support, they hide the FP8 under the BF16 model group in the side menu...
Early Access is enabled to support the development of new versions and models.
Description
FAQ
Comments (28)
So... why do you still develop this? Flux 1 is old tech, and wasn't even good to begin with. The latest version's previews still look as plasticky and fake as Flux did on release back in 2024 or whenever it came out.
I mean, I'm more curious to why you would go around commenting stuff like this on models? What's the benefit to you?
For me, I like this model and I think it looks great, so I'm continuing developing it.
I'm still working with SDXL because I think it's fun and beautiful.
@6tZ Sorry, I'm just shocked people like it. Also want to understand why the community seems to stick to certain models and abandon other ones.
I see the value with SDXL, I get why people use that one even now. I just don't think see the value prop for flux 1 though
@sircringealot @6tZ I don't understand why everyone is attacking the guy, he was just asking. I completely agree with his opinion.
@iwid224 Of course everyone is entitled to their opinion. I didn't attack anyone, I asked a follow-up question, while I also replied to their question.
@sircringealot I don't think this is a constructive way of figuring this out. If you don't see the value of this architecture, that's fine. I explained why I do it. I think it's a great model. Much stronger than SDXL. Of course other more modern models outrank it by now, but it's still a good foundation.
@sircringealot Oh, and we are good btw, no animosity or hurt feelings here <3
I just don't think we need a reason to make beautiful models.
@6tZ I think hes saying that because you could use your talents for a newer lora on a new, better model, like ltx2.3 or klein9B or zimagebase.
@lolmao500 Fair. Still there are better ways of asking for that rather than insulting the quality of the model and shit-talking it the model and architecture. It's an opinion that OP is allowed to have of course, but it's not a great way of talking to people that puts hours and money in to train something and give to the community.
From what I've read, it seems like Flux.2 has proven more difficult to modify in these specific ways. Something about not being a proper base model and not being worth the trouble. I don't know enough to say how true any of that is. I think it does fix some of the issues with Flux.1, but I also think there are ways of working around those problems to some extent.
@wawecis961781 People talk shit all over the internet. Basically don't believe anything you read and just make up your own mind if you have time and care enough.
@6tZ Yes. Your checkpoint is very nice so please don't let them get to you. I think my only problems are with the CLIP, so I've been experimenting with different models and reading about how they work. Still a very long road. 🙂
@wawecis961781 Thanks kindly.
Yeah there's so much to learn. But once you get into training, it's damn fun!
@6tZ I would definitely like to try training at some point. Hopefully it won't take too long on my one 9070 XT. My computer doesn't have enough PCIe lanes for another card and the official requirements listed for training on some of these models are very intimidating (90+GB of V.RAM). 😅
@wawecis961781 That's for full fine-tunes.
You can train a LoRA on much lower requirements, and then merge this into the checkpoint.
The results aren't the same, but it works. This is how I train my models.
Is there a chance to see one of your models base on flux schnell?
Probably won't do that.
@6tZ Chroma is based on schnell and it's already pretty good at adherence, NSFW and photorealism. it's underrated for sure, probably just needs some reps from creators such as yourself to bring it up a couple notches. please give it a look if you haven't already.
@dow185 That's true.
I did play around with it briefly, but it wasn't that good for me. I didn't give it enough of a chance though. So many models and formats to try. Can't commit to all of them :/
Chroma, Ernie, Anima are some of the top contenders still.
@6tZ Chroma seems like the one with the most potential imo, it's already really quite good maybe check out the Gonzalomo model. what's interesting about Chroma is it's compatible with all flux1 bases... I saw someone merged it with Krea for example. I didn't try it bc it was 40GB+ XD
at the very least you'd probably see gains in NSFW and maybe adherence for your flux.d models if you merged with Chroma.
Thank you as always for the wonderful models. What are the differences between this model and Fux Capacity? Which one would you recommend?
I recommend the one you like the best aesthetically. That's the one that I think is a perfect match for you. But this is strictly between you and me, and everyone else reading this comment. So keep this to yourself. It'll be our little secret, yeah?
The difference is that they are different models and produce different results. So one looks different, while the other one looks distinctly alternatively.
The trick here, is that it's all subjective, and it doesn't matter, so just go with either, or both, or neither, it's all good!
Will it work well with img2img ? I would like to use my own model
I don't understand the question.
@6tZ Like am I able to edit my own images using this model? Am i able to make myself nude.
@loloboo27421 Ah, hmm, I would suggest the Flux2K models for that:
https://civitai.red/models/2547526?modelVersionId=2913315
https://civitai.red/models/2453960/miraclein-nsfw-generation-and-edit-flux2klein
These are called "Edit Models", with the capacity to take an instructional prompt, and update the image based on that.
You can use this for img2img, but it's more precautious. I would use those instead.
@6tZ thank you
@6tZ I would just like to point out that this could be the single most helpful response on the entire homepage. Bless! <3