[Stable Diffusion] Prompt Sharing and Learning Thread

[Synalon]

Z IMAGE.
It does ok at higher res, I find it wants more long form descriptive prompts, instead of just keywords or tags.
It will improve with some loras for better nudity but it is a decent start right now.

 

[Synalon]

This is a couple of screenshots from the Danbooru website to help with tags for " T Girl" and beastiality.

For beastiality I would assume you need to add the name of the animal within the brackets shown in the example prompt in the image.


Mating (DOG), would be my assumption, along with the sexual position, such as "on all fours". It's not my kind of thing, so I've never looked into it though.
The list of animals is quite large, so I do recommend looking on the Danbooru site to see what should be available.

 

[Sharinel]

I usually use Illustrious, since I want to dedicate myself to creating anime-style images; I've tried with animal Lora, but they are very scarce.

Poor old Lara gets it from everything

Panther


Minotaur


Tiger


Honestly though? You don't need a Lora - just put in 1boy, german shephard, doggystyle, canine penis. Boom

 

[devilkkw]

Z-image, 1400x1800 native:


The image is done in 4 step at cfg 1.8.
I'm amazed with this model.

 

[Synalon]

Z Image does text pretty well, I haven't had any issues with the text yet at least.

 

[Monki0]

Z-image, 1400x1800 native:
View attachment 5483040 View attachment 5483038 View attachment 5483037 View attachment 5483034

The image is done in 4 step at cfg 1.8.
I'm amazed with this model.

Would you say Z Image does photorealistic NSFW well? Or is Pony a better pick here?

 

[Sepheyer]

...

Fren, maybe it is time to get AI to get you a new logo? I mean it in a supportive way ^_^

It is a cool lil project to spend an evening to see what you can generate that you don't totally hate ^_^

This is mine using Qwen 2509 (I just tried using ZT with the same prompt and I hated it):

 

[Sharinel]

Would you say Z Image does photorealistic NSFW well? Or is Pony a better pick here?

Z image is a 'base' model, it's not been tweaked to have a lot of NSFW. I'm sure those will come out in time but at the moment it's not brilliant for anything except topless. Pony is a NSFW model, so if that's what you are interested in then Pony would be the better bet (although illustrious and NAI are better than Pony imo)

 

[JhonLui]

Z image is a 'base' model, it's not been tweaked to have a lot of NSFW. I'm sure those will come out in time but at the moment it's not brilliant for anything except topless. Pony is a NSFW model, so if that's what you are interested in then Pony would be the better bet (although illustrious and NAI are better than Pony imo)

Give ZImg. sometimes, even Illustrious took more than a year to get some decent realistic Nsfw, but now it's king of the hill.

 

[devilkkw]

Would you say Z Image does photorealistic NSFW well? Or is Pony a better pick here?

I hope

Spoiler: this

You don't have permission to view the spoiler content. Log in or register now.

answer your question. Also read at the end of this postfor some trick.

Fren, maybe it is time to get AI to get you a new logo? I mean it in a supportive way ^_^

It is a cool lil project to spend an evening to see what you can generate that you don't totally hate ^_^

This is mine using Qwen 2509 (I just tried using ZT with the same prompt and I hated it):

View attachment 5484631

Interesting for try. I give it a chance. Also use rembg to remove bg

Some trick for ZIT you have to keep in mind:
After 2 days of test, using default workflow, this is what i discovered:

Enclose your prompt with <think> PROMPT HERE </think> give better result's, especially in fantasy character's

Also, re-enablig model sample aura flow node (bybassed by default) and set it to 5, give better result's in anatomy, but image lose some detail's.

Someone said use system prompt give better result's, but never tested. THere's a

You must be registered to see the links

for doing it, i will test soon.

 

[Thaneas]

Hmm, I'm still relatively new to this whole ai image generation thingy, and I have encountered a roadblock, so please forgive my noobness and bear with me. I'm trying to make a hentai-style pic that depicts a girl who got raped while crying and have one of her arms covering her eyes.


Ref pic above

I've already tried prompts such as one arm over eyes, one arm on eyes, covering face with arm, but still no dice. Tried browsing danbooru for relevant tags, sadly they're focusing more on hands, not arms, so I'm stuck here. Any suggestion to resolve this roadblock? Thank you in advance

 

[JhonLui]

Ref pic above

Please take this advice lightly, since I'm pretty scarse in prompting specially for videos.

First you should specify what Program - Checkpoint - Vae you are using, since there is a difference how they react to prompts...

Regarding the prompt: this was the best I could get with : "right arm over eyes, (closed hand), eyes covered,"



Tried a ton of solutions, including yours, but were all worse.
Luckly I remember there is an Illustrious Lora on Civitai depicting exactly that situation, look for "forced missionary" or similar.

Many programs like A111 and Fooocus have a "Describe" function, so if you load that image, they will elaborate that prompt, or you can simply work Image to Image and change the subject with what you neeed.

Also installing and learn to use control.net can help a lot (and I should do that too).

 

[Thaneas]

Please take this advice lightly, since I'm pretty scarse in prompting specially for videos.

First you should specify what Program - Checkpoint - Vae you are using, since there is a difference how they react to prompts...

Regarding the prompt: this was the best I could get with : "right arm over eyes, (closed hand), eyes covered,"

View attachment 5512346

Tried a ton of solutions, including yours, but were all worse.
Luckly I remember there is an Illustrious Lora on Civitai depicting exactly that situation, look for "forced missionary" or similar.

Many programs like A111 and Fooocus have a "Describe" function, so if you load that image, they will elaborate that prompt, or you can simply work Image to Image and change the subject with what you neeed.

Also installing and learn to use control.net can help a lot (and I should do that too).

Thank you for the response. Truth to be told, i'm still testing the waters by using free online ai generation tool so i'm not in total control of the software. Though said online tool does have models and loras selection. I'll browse around Civitai as you suggested to see if there's workaround to solve this.

 

[JhonLui]

Thank you for the response. Truth to be told, i'm still testing the waters by using free online ai generation tool so i'm not in total control of the software. Though said online tool does have models and loras selection. I'll browse around Civitai as you suggested to see if there's workaround to solve this.

If you have an SSD to dedicate and a decent GPU (rtx3060 up) you might want to look for Stability Matrix.
Can't say more or the post will be banned for advertisement... but it's worth a check.

 

[devilkkw]

i've made a copy/paste workflow for Z-image Turbo, it's made for add detail's and variation without any extra custom node.

You must be registered to see the links

Just replace your sampler with this. Read detail's on civitai.

 

[Sepheyer]

i've made a copy/paste workflow for Z-image Turbo, it's made for add detail's and variation without any extra custom node.

You must be registered to see the links

Just replace your sampler with this. Read detail's on civitai.

Can you post a few pics to give an idea of what it does please?

 

[devilkkw]

Can you post a few pics to give an idea of what it does please?

All image are in the civitai page, i've uploaded some gallery. Is not visible? let me know and i reupload here.

 

[JhonLui]

All image are in the civitai page, i've uploaded some gallery. Is not visible? let me know and i reupload here.

They are visible allright, and look beautiful.
Only complain is that I didn't get the girl when I bought that bike!

 

[devilkkw]

They are visible allright, and look beautiful.
Only complain is that I didn't get the girl when I bought that bike!

LOL. you need to contact Honda Customer Center for solving the issue

 

[HinsdaleDays]

Appreciate this thread, thanks to all who have contributed. I'm releasing a fairly massive Twine game soon-ish that uses AI images, as a necessary evil, despite my dislike of most all things AI. This thread definitely helped some.