Photos & Videos AI Tools: How to and Technical Discussion on Creating NSFW Gay Sex Stuff with Stable Diffusion (and others)

[StudmeatFL]

Starting up a new thread for those folks wanting to learn more about the Tools, Models, Methods and How-Tos for creating consistently hot NSFW AI images of MEN with Stable Diffusion and other tools. There are plenty of tutorials out there for women, but very few focusing on Men, gay sex, etc.

Please feel free to post tutorials, your own step by steps and reference links along with example images. If you're great at creating consistent prompts, feel free to add those, too.

Some Useful Links to start:
Civitai: for NSFW Models, Checkpoints and Lora's (I direct linked to AirFuckBrute's Mix:
Airfuck's Brute Mix - v1.0 | Stable Diffusion Checkpoint | Civitai
- A generic guide for generating NSFW with Stable Diffusion:
Stable Diffusion NSFW - Images, Prompts Models in Google Colab - Cloudbooklet

 

[StudmeatFL]

Another Free Tool if you want to try to run locally on a Mac. It does require an Apple Silicon Mac (M1, M2, M3, etc), but the added benefit is that it runs locally so all your images are private:
DiffusionBee - Stable Diffusion App for AI Art

 

[Fjsuhwr]

xpost from the previous thread - how to use SD without a PC/laptop

sign up here: https://tensor.art/images/712535220...0710122895&source_id=nj-0pl_hlEG0rfosb3f28RUg

go here: Virile Stallion - V5.0 Photo | Stable Diffusion Model - Checkpoint

some setting:

Press |> Run;

The author recommends the following parameters. Do you want to apply them? - press Use Recommended

Prompt:
score_9, score_8_up, score_7_up,
cinematic, realistic, 1boy, muscular male

Negative prompt:
low quality, lowres, bad anatomy, normal quality, worst quality, 3D, render, illustration, drawing, comic, watermark, greyscale, monochrome

Follow these settings, have fun!

it'll take time to load and u shud be able to see ur creation there

 

[Fjsuhwr]

some notable checkpoints

Virile Reality - V3.0 BETA 3 | Stable Diffusion Checkpoint | Civitai

Virile Stallion - v5.0 Photoreal | Stable Diffusion Checkpoint | Civitai

 

[StudmeatFL]

some notable checkpoints

Virile Reality - V3.0 BETA 3 | Stable Diffusion Checkpoint | Civitai

Virile Stallion - v5.0 Photoreal | Stable Diffusion Checkpoint | Civitai

Not too bad surprisingly useful tool and nice to get into SDXL arena. First image was just from my prompt:
naked muscular male kneeling outdoors by a swimming pool, looking over right shoulder directly at viewer, right hand on ass, view from behind, back arched, legs spread wide, short black buzzed haircut, 40yo, mouth open in ecstasy, huge uncut erect penis, covered in cum, very sweaty, full body view

Then decided to do an Img2Img using a reference image and added some additional prompting:
naked muscular male kneeling outdoors by a swimming pool, looking over right shoulder directly at viewer, right hand on ass, view from behind, back arched, legs spread wide, short black buzzed haircut, 40yo, mouth open in ecstasy, huge uncut erect penis, covered in cum, very sweaty, full body view, multiple cocks in foreground ejaculating onto subject, cumshots

 

[Fjsuhwr]

there should also be a discussion for controlnet!

 

[StudmeatFL]

there should also be a discussion for controlnet!

we can add it to this thread. I think discussion of methods works pretty good here. I'd like to find some good face replacement myself.

 

[StudmeatFL]

Not too bad surprisingly useful tool and nice to get into SDXL arena. First image was just from my prompt:
naked muscular male kneeling outdoors by a swimming pool, looking over right shoulder directly at viewer, right hand on ass, view from behind, back arched, legs spread wide, short black buzzed haircut, 40yo, mouth open in ecstasy, huge uncut erect penis, covered in cum, very sweaty, full body view

View attachment 138671781

Then decided to do an Img2Img using a reference image and added some additional prompting:
naked muscular male kneeling outdoors by a swimming pool, looking over right shoulder directly at viewer, right hand on ass, view from behind, back arched, legs spread wide, short black buzzed haircut, 40yo, mouth open in ecstasy, huge uncut erect penis, covered in cum, very sweaty, full body view, multiple cocks in foreground ejaculating onto subject, cumshots

View attachment 138671601

Managed to make them more manly and beefier and get some hair on that body

 

[StudmeatFL]

xpost from the previous thread - how to use SD without a PC/laptop

sign up here: https://tensor.art/images/712535220...0710122895&source_id=nj-0pl_hlEG0rfosb3f28RUg

go here: Virile Stallion - V5.0 Photo | Stable Diffusion Model - Checkpoint

some setting:

Press |> Run;

The author recommends the following parameters. Do you want to apply them? - press Use Recommended

Prompt:
score_9, score_8_up, score_7_up,
cinematic, realistic, 1boy, muscular male

Negative prompt:
low quality, lowres, bad anatomy, normal quality, worst quality, 3D, render, illustration, drawing, comic, watermark, greyscale, monochrome

Follow these settings, have fun!

it'll take time to load and u shud be able to see ur creation there

So far I've been extremely impressed with tensor.art's offering, even without paying.

I definitely prefer it over Google Collab. Do you know if they offer any kind of Re-Actor of face replacement/face swapping? I tried the training Lora but it was pretty crap.

 

[CanesellerX]

Omg this is life saving. It's so hard to find ai that can generate proper penis.
Also I have question! Is there any app compatible for android that can generate hoy guy?
(I'm totally just dump horny man who doesn't know anything about ai at all)

 

[StudmeatFL]

Omg this is life saving. It's so hard to find ai that can generate proper penis.
Also I have question! Is there any app compatible for android that can generate hoy guy?
(I'm totally just dump horny man who doesn't know anything about ai at all)

You would want to go the route of tensor.art, since it's a free login and you get 100 credits a day. Follow that tutorial the fellow posted above with the direct link. It will work web based, but they also have an android app.

 

[StudmeatFL]

If it helps any, here's a sample generation that got me started in the direction I was wanting for a Gloryhole sequence.

##--- Copy everything from here down to the END ----

Prompt: double gloryhole suck and fuck, naked man on all fours, naked, muscular, hairy, beefy, short black hair, looking directly at camera, full body view, huge erect uncut cock, mouth open in ecstasy,
Negative prompt: EasyNegative, shoes or clothing,
Steps: 35,
Sampler: DPM++ 2M SDE Heun Karras,
KSampler: dpmpp_2m_sde_gpu,
Schedule: karras,
CFG scale: 7,
Seed: 0,
Size: 1152x768,
VAE: sdxl-vae-fp16-fix.safetensors,
Denoising strength: 0,
Clip skip: 2,
Model: VirileStallionPRv2.fp16

## -- END ----
(don't copy the start and end stuff, just the stuff from Prompt down to Model)

When you're over at Tensor.Art (or logged in and click 'Creator' at the top left there's a block for 'Paste Generation Data'. Click that and allow the website to access your clipboard. It will pull it in and populate it into the Creation screen. See gif below

I was out of credits at this point (they refill at 5am local time) but this is how simple it is to populate. Take a little time to get to know what the different sections are.

When I ran that, it took a few minutes and generated a set of 4 images. I picked the one I liked out of the set that would be a good starting point and clicked 'Remix':

That gave me another great set of images (ran it twice):

Then I sent it over to 'Send to img2img' which takes your current photo and uses it as a reference. Alternatively, you can download the image you like and upload it directly into img2img. Set the 'Denoising Strength closer to 0 if you want something close to the current image and closer to 1 if you want more variations. I used a 0.3/0.4 I believe to get this image:

## -- START

Prompt: naked muscular beefy man being fucked from behind and sucking cock through a gloryhole, side view, cock penetrating_ass, anal sex, gay sex, gloryhole sex, blowjob, cock_ring, huge erect uncut cock, looking_at_viewer, full body view, very sweaty, hairy legs, hairy chest, short black haircut, cumshot, thick_cum, mouth open in ecstasy, full body view,
Negative prompt: EasyNegative, no shoes or clothing,
Steps: 35,
Sampler: DPM++ 2M SDE Karras,
KSampler: dpmpp_2m_sde_gpu,
Schedule: karras,
CFG scale: 7,
Seed: 0,
Size: 1152x768,
VAE: sdxl-vae-fp16-fix.safetensors,
Denoising strength: 0.25,
Clip skip: 2,
Model: VirileStallionPRv2.fp16,
ADetailer model: face_yolov8s.pt,
ADetailer prompt: handsome muscular face, open eyes, mouth open in ecstasy, tongue out,
ADetailer negative prompt: ,
ADetailer confidence: 0.5,
ADetailer dilate/erode: 4,
ADetailer mask blur: 4,
ADetailer denoising strength: 0.5,
ADetailer inpaint only masked: true,
ADetailer inpaint padding: 32

## -- END

ADetailer is used to fix faces, typically, but you can also use it to just focus on the face and change things about the face (like facial expressions).

ControlNet gives you more power over "controlling" your pose, or making the current image closer to the existing one. This is great if you have an image that's a pose you like, but you want to change up the scenery, etc.

So in a nutshell:

1. Model defines the overall look and feel of what you're going to be able to create, the Virile Stallion model gives you MEN and cocks, etc.
2. Prompt lets you define what you're trying to create. Still learning this but take a look at images you like and you can pull from their prompt
3. Steps determines how many runs through it goes to build your image. Keep this between 25-35, I believe.
4. Add ControlNet or Lora to further define your image. I've Added Lora for 'Cum/Bukkake' to get a whole bunch of cum in my final image before. ControlNet Pose copies a stick figure pose, Depth takes an existing image and tries to grab the overall look of it. I prefer Depth over OpenPose.
5. Adetailer fixes faces. I think it can also fix hands and feet, but I don't see those offered in Tensor.Art.

Experiment and have fun!

 

[Fjsuhwr]

So far I've been extremely impressed with tensor.art's offering, even without paying.

I definitely prefer it over Google Collab. Do you know if they offer any kind of Re-Actor of face replacement/face swapping? I tried the training Lora but it was pretty crap.

LORA training i do it on my own laptop, it's not too bad actually as long as u have good quality data with clear captions

 

[StudmeatFL]

LORA training i do it on my own laptop, it's not too bad actually as long as u have good quality data with clear captions

But how do you get the Lora available to use in your Tensor.art stuff. Do you have to publish it in a project and make it public?

 

[Fjsuhwr]

that i dont have the answer to, i dont really use tensor.art to generate stuff - i found it as an alt solution for others who doesnt really have a strong equipment to generate imgs

 

[StudmeatFL]

that i dont have the answer to, i dont really use tensor.art to generate stuff - i found it as an alt solution for others who doesnt really have a strong equipment to generate imgs

I gotcha, that makes sense. I have a Mac, so Stable Diffusion locally isn't really an option.

 

[snowchyld]

Thanks for the tutorial. This is really fun and easy.

 

[StudmeatFL]

Thanks for the tutorial. This is really fun and easy.

Awesome! It really is not that complicated... well it's complicated but the complicated parts are the other settings. Most of those you set and forget and then focus on the prompts.

 

[Im23XXzz]

How can I inpaint to make someone in a picture nude? I'm using DiffusionBee and I downloaded the Homo Fidelis model (HomoFidelis - v4.0 | Stable Diffusion Checkpoint | Civitai) and clicked the Inpainting in the app, made a mask around the underwear of the model. And as a prompt I tried things like, nude, cock, dick, penis. But no matter what I do it always either puts a pussy there (instead of a cock) or just flesh. I did import and select the Homo Fidelis model.

How do I get it to paint a cock instead of a pussy?

 

[StudmeatFL]

How can I inpaint to make someone in a picture nude? I'm using DiffusionBee and I downloaded the Homo Fidelis model (HomoFidelis - v4.0 | Stable Diffusion Checkpoint | Civitai) and clicked the Inpainting in the app, made a mask around the underwear of the model. And as a prompt I tried things like, nude, cock, dick, penis. But no matter what I do it always either puts a pussy there (instead of a cock) or just flesh. I did import and select the Homo Fidelis model.

How do I get it to paint a cock instead of a pussy?

I’m not seeing any cocks in that particular model, at least not in the examples, though it was based on some that would be able to handle it. I’ve always used Airfuck's Brute Mix - v1.0 | Stable Diffusion Checkpoint | Civitai and it’s done fine, but I’ve also never experimented with inpainting yet.