This is the Pulsar checkpoint.
A male focused Nude/NSFW model.
This is the the starting point on a male-focused model, but it will also be used to merge with FuxCapacity to hopefully unlock full male + female NSFW and porn checkpoints.
One step at a time, this is already producing solid male nudes so I wanted to share it.
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.
The model is still a work in progress, and the focus is on adding penis gentalia to Flux.
It should support additional poses like from behind, anuses and masturbation, penis POV.
There is captioning support for controlling the penis erect or flacid state, but Flux did not learn it properly.
The model should already allow for a range of ethnicities. Try prompting with a random nationality.
Future training will focus on including:
Hairy bodies
Male to male interactions (homo erotic at least)
[Your feedback / requests below]
Description
FAQ
Comments (37)
It looks very cool what you achieve with it. Male anatomy is hard for every model! Thanks.
But it does feels very pony/plastic like, which I personally don't like.
Thanks!
Yeah I agree on that. The plasticness is from base Flux I think. You need to prompt very hard to get around that. Possibly it could be trained out as well, but I'm not confident about that :)
would love controllable body hair, and also controllable pectoral size.
Yeah, body hair would be a target for the next version.
Once it's figured out how to tame flux...
@6tZ ya, especially controllable. Like, smooth / shaved / short / curly / long / thick.
Maybe training over the undistilled or FluxDev pro (or both, like the nude humans checkpoint) would help?
@Janet Trying that right now. Not sure if it's better or not yet.
Really nice and hard fun but it's looking more like Pony, bodies aren't as "fluxable"?
Can you tell me more of what you mean by it looking more like pony, and bodies aren't fluxable?
@6tZ Before applying the LORA the hot man looks ultra realistic like FLUX in general but if I apply the LORA even 0.2 the man becomes less realistic and their faces specifically become dull and shiny like Pony models. I love the work you're doing for the dong community though so appreciate the fun testing!
@KornLover Right, gotcha. You mean the other model, right? Not this checkpoint, which isn't a LoRA?
But yeah, it's trained on pony and SDXL images, so it makes sense what you're saying.
I was just feeling that the images from this model felt more Flux than Pony.
it's pretty good, but all the bodies come out too muscular and hegemonic, even indicating skinny, fat, small or twink, the result is always muscles everywhere, if you can fix that, this would be a great model!
You're right by default, you have to really prompt for it.
It can do fat or bulky. But it doesn't do skinny very well.
There's a skinny body LoRA available though which could be used.
I'll add this to the "todo" for the model.
very good thanks, as you said one ste at the time
Love this model, I hope you keep improving it.
Thank you!
I plan on doing so.
Thanks for the great work! 💯 This looks promising.
I'm attempting to train LoRAs using this but I'm having inconsistent results while training it on this pruned model. There's not much Flux models which can handle male genitals so it'll be awesome if you have a base model. Thanks again!
I only saved this out as fp8, I don't find too much of a difference with higher fidelity, that's not the problems with progress on these models yet at least.
Maybe you can merge it into an fp16 or bf16 or whatever you want to use? Or re-save it as diffusers somehow. I don't have tools for this :)
@6tZ Sure, I'll look into it. Thanks! 🙂
Thank you for making male-oriented AI. It is often hard to find! <3
Happy you like it!
I was looking to get Flux to handle both men and women at the same time. So male dataset creation was needed to proceed and I figured I'd make the most out of it ^^
Now... back to all the mussies =_=
Hi there! ✨
First off, amazing work on the Pulsar model—it’s really exciting to see a male-focused NSFW base like this. 🖤 I’ve been trying to test it out in ForgeUI, but I ran into an issue: I’m getting an error saying, “You do not have CLIP state dict!” when trying to generate images.
Is the CLIP component included in this version of the model, or is it expected to be added manually? If it’s a known issue, do you have any suggestions for resolving it? I’d love to fully utilize this model.
Thanks so much for your hard work and for sharing this with the community! 🙌 Looking forward to your feedback. 💖
Hi. Thanks for the kind words.
Unfortunately, I have no idea about Forge. It doesn't seem stable at all for Flux models. I've been trying to make sure my models work for both Forge and ComfyUI, but I'm not able to successfully use models I'm saving in Forge with either tool, so I have no idea how to solve this problem right now.
Unfortunately my only recommendation for you is to try ComfyUI or a tool that uses it as a base.
Sorry for the trouble!
@6tZ Hi, it actually does work. It's just that the files are not included in your checkpoint, and i'm not that experienced. But i added these files myself and selected them in the VAE / Text Encoder dropdown. Forge UI has been better than A1111 since it's not updated anymore to my knowledge.
I'm still scared to try ComfyUI, I'm not really there yet in terms of experience ahah.
Thank you for the response! Merry Xmas :3
@Seunae Ah good to know.
Yeah Forge was great, until Flux came out, and now it feels very unreliable IMO.
Happy it's resolved, and now if people ask me I'll just tell them to try loading VAE/Encoder separately.
Also I'll see about including them in my model, not sure which is better.
@6tZ Makes sense why some creators don’t include those files—it saves storage, and you can reuse them across checkpoints. Some even list the necessary files in the description, which helps a lot. But yeah, adding them yourself works great once you know! Happy holidays! :3
@Seunae What files were those? I am having the same issue you are on Forge :(
@serpentine
You can find the official files here:
https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main
The ae.safetensors in the root of that folder:
https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/ae.safetensors
And inside text-encoder:
https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main/text_encoder
You have a model.safetensors-file which should be the encoder you need.
@serpentine
1: ae.safetensors
2. clip_l.safetensors
3. t5xxl_fp8_e4m3fn.safetensors
i have all those three in my text_encoder folder. i think OP explains well with links in his response.
@StoneFruitAI put aa.safetensors in VAE folder, this one should't be in text encoder folder, other two files are fine
i"am sorry
Only the SDXL and the best in this field and don't tell me the opposite? gives a single photo with its data as proof. with better quality???
Can someone pls provide me the workflow, idk how to get it working
Each image uploaded on the model (at least by me), should have a ComfyUI workflow included. You will need to update the models so they match your system, and optionally replace some nodes that aren't fully necessary. Like the "Save image with metadata" node etc.
Hi all :)
First of all thanks for the efforts training this model!
Second, could anyone maybe provide me with a workflow for Flux that's capable of handling this model?
I kept struggling for hours with a standard Flux workflow, altering it until failing either with "attention_mask_img_shape" problems or some other errors in the final node before image output.
Thanks in advance 🙌🏻
Sure, what GPU are you on? How many gb VRAM?
Omg this on-site would be amazing 😍
Yeah, you need to use the website's "Bid" process to vote for which models should be included on the generator each week. It's about 10000 buzz per week currently.
Please consider making a Flux.2 Klein 9B version of this checkpoint.