Renderotica

Hey everyone! 👋

I have some incredibly exciting news to share. Today was the day—I finally generated my very first image that is completely AI-driven, from start to finish! Not just the picture itself, but also the prompt that created it (which I only gave a rough idea for!).
Not that this is anything particularly new for AI professionals. But for me, it's the first time.

Honestly, the sheer amazement of what you can achieve with such simple tools and without needing any expensive online subscription is mind-blowing.

This is just the starting point, and I know it is possible to push the detail and artistic flair way up—with AI help or with a human brain! 

Here’s how the magic happened:

The Setup: LM Studio (seriously, one-click installation!).
The Brain: Downloaded an uncensored LLM—specifically, Gemma.4.E4B-Uncensored.
The Process: Asked the model for a prompt based on my rough idea.
The Art: Took that perfect prompt and fed it into Stability Matrix to generate the image using Z-Image!

The hardware: Nothing special. I even got the LLM (for the prompts) running on an old laptop.
For generating images, I use a fairly standard machine with a GPU. Anything with 8 gigabytes of VRAM or more will do the job. More is recommended; things get interesting starting at 16 gigabytes.

The most interesting thing about this story: There’s no language barrier. You can use your native language to write the prompts. The AI—or rather, the LLM—knows how to work with Stable Diffusion and knows all the nuances involved in extracting specific details from the images.

Furthermore, there’s no censorship by the AI. And I mean: NO CENSORSHIP!

Originally Posted by: dsv4600

I don't understand this process. Well I do, but I don't understand your comment about there being no online subscription. The last time I looked, Z-Image was a subscription service ($6.99/month). Are they now allowing you to generate uncensored pornographic images for free without being constrained by their terms and conditions? If so I'd have thought it was better to iteratively generate your own prompts and enjoy the sense of satisfaction that comes from getting a bloody AI to actually create the image you asked it to create. Interesting post, though. Have fun!

You can run all that on your local computer. All software is open source. Just the cost of a download :)

https://github.com/LykosAI/StabilityMatrix

This is a central piece of software. It lets you manage different software products and all the models you need.

Here you can find the models: https://civitai.red/

(You have to register to see NSFW content – it's also free to use!)

If you use the software and models offline, no one will be able to control or censor you anymore. No matter what your desires may be.

Originally Posted by: ADP001

Originally Posted by: dsv4600

I don't understand this process. Well I do, but I don't understand your comment about there being no online subscription. The last time I looked, Z-Image was a subscription service ($6.99/month). Are they now allowing you to generate uncensored pornographic images for free without being constrained by their terms and conditions? If so I'd have thought it was better to iteratively generate your own prompts and enjoy the sense of satisfaction that comes from getting a bloody AI to actually create the image you asked it to create. Interesting post, though. Have fun!

You can run all that on your local computer. All software is open source. Just the cost of a download :)

https://github.com/LykosAI/StabilityMatrix

This is a central piece of software. It lets you manage different software products and all the models you need.

Here you can find the models: https://civitai.red/

(You have to register to see NSFW content – it's also free to use!)

If you use the software and models offline, no one will be able to control or censor you anymore. No matter what your desires may be.

Thanks for those links. I might have a play as I've pretty much given up on generating NSFW on any of the mainstream platforms because they are now so censored and locked down, not to mention the costs.

Originally Posted by: ADP001

Hey everyone! 👋

I have some incredibly exciting news to share. Today was the day—I finally generated my very first image that is completely AI-driven, from start to finish! Not just the picture itself, but also the prompt that created it (which I only gave a rough idea for!).
Not that this is anything particularly new for AI professionals. But for me, it's the first time.

Honestly, the sheer amazement of what you can achieve with such simple tools and without needing any expensive online subscription is mind-blowing.

This is just the starting point, and I know it is possible to push the detail and artistic flair way up—with AI help or with a human brain! 

Here’s how the magic happened:

The Setup: LM Studio (seriously, one-click installation!).
The Brain: Downloaded an uncensored LLM—specifically, Gemma.4.E4B-Uncensored.
The Process: Asked the model for a prompt based on my rough idea.
The Art: Took that perfect prompt and fed it into Stability Matrix to generate the image using Z-Image!

The hardware: Nothing special. I even got the LLM (for the prompts) running on an old laptop.
For generating images, I use a fairly standard machine with a GPU. Anything with 8 gigabytes of VRAM or more will do the job. More is recommended; things get interesting starting at 16 gigabytes.

The most interesting thing about this story: There’s no language barrier. You can use your native language to write the prompts. The AI—or rather, the LLM—knows how to work with Stable Diffusion and knows all the nuances involved in extracting specific details from the images.

Furthermore, there’s no censorship by the AI. And I mean: NO CENSORSHIP!

 

 

I love to work with ComfyUI, although I do it in the cloud, since my laptop doesn't have the power I'd need to run it locally.
What's the requirements to run  Stability Matrix locally? I guess it also depends on the Models you're using.

Thanks

Originally Posted by: dsv4600

Damn that actually works! I'll admit to being skeptical but 40GB of downloads later I had my first image out of the setup. I skipped the business with LM Studio making a refined prompt as I've done quite a few AI prompts and prefer to refine them myself - though I might use that LLM for other things. Thanks again for all the info!

Congratulations! I hope this thread will encourage even more people to skip the online sites if they have the necessary hardware. After all, many of these online services do nothing but take open-source software and free models and then offer them with severe limitations for a lot of money.

An LMM can provide a good starting point for a topic when working with prompts. 
What AI is particularly good at is taking into account all the subtle details that can make an image great. Sometimes, it’s just a single added word that makes all the difference.

What I especially like is that you can discuss the prompt (and thus the specific image theme) with the AI. It can even go as far as: “Let’s try this and test that.” If you use an uncensored LLM (as described above), you’ll find—perhaps to your surprise—that the AI can be pretty raunchy :)

Tell us what software you're using. Maybe someone who already has experience with local installations can help you (and others) out.

Edited by user Sunday, August 2, 2026 5:12:58 AM(UTC)  | Reason: Not specified

Originally Posted by: ADP001

Originally Posted by: dsv4600

Damn that actually works! I'll admit to being skeptical but 40GB of downloads later I had my first image out of the setup. I skipped the business with LM Studio making a refined prompt as I've done quite a few AI prompts and prefer to refine them myself - though I might use that LLM for other things. Thanks again for all the info!

Congratulations! I hope this thread will encourage even more people to skip the online sites if they have the necessary hardware. After all, many of these online services do nothing but take open-source software and free models and then offer them with severe limitations for a lot of money.

An LMM can provide a good starting point for a topic when working with prompts. 
What AI is particularly good at is taking into account all the subtle details that can make an image great. Sometimes, it’s just a single added word that makes all the difference.

What I especially like is that you can discuss the prompt (and thus the specific image theme) with the AI. It can even go as far as: “Let’s try this and test that.” If you use an uncensored LLM (as described above), you’ll find—perhaps to your surprise—that the AI can be pretty raunchy :)

Tell us what software you're using. Maybe someone who already has experience with local installations can help you (and others) out.

Actually I'm just using the software you mentioned in your first post, namely Stability Matrix with the ComfyUI plugin and the Z-Image models and it's all working fine. I've since enhanced that by installing the ComfyUI Manager (no easy task actually) and that makes it easier to download new models and nodes from within the local ComfyUI.

The graph view threw me for a moment but as soon as I saw what was going on it was pretty easy to use and I tested several AI prompts from previous experiments with online sites. Very good quality and as you said, you are in total control of what the model produces.

I run an RTX 4070 which is pretty mid range nowadays but its 12GB of VRAM is more than enough to handle the task in typically 20 seconds although I haven't tried it with super hires images yet. The downside is the staggering amount of hard disk space that you need to store the models but that's better than $6.99/month for a censored "we know what's best for you" service.

All that said I would never take AI over 3D image creation where you have absolute and very fine control over the output.

Thanks.

Originally Posted by: dsv4600

Blender suffers from a "Do Everything" mentality that has led to a user interface of mind blowing complexity.

Exactly. and ComFy is AI's Blender.
Especially since Blender has a node-system – like ComFy :)

Originally Posted by: dsv4600

I do think the term "Artifical Intelligence" is a big misnomer - this software is not "intelligent" it is not capable of independent thought in any meaningful way, at least not yet.

Maybe it's just a misunderstanding.
Every form of intelligence requires training in a specific field. This is especially true for “artificial” intelligence. 
If the training is focused on specific areas, the intelligence can only excel in those areas. If certain things weren’t taught to it—or were taught incorrectly—that’s not a problem with the intelligence itself, but rather a problem with the trainer.

It’s certainly true that marketing has fueled quite a few false expectations among users—often due to a lack of understanding on the part of advertisers.

I use an LLM for generative AI mainly because the LLM knows the specific syntax used to train the models. Just yesterday, I encountered a case where a Z-Image model generated poster-like images instead of photographs on a specific topic. By interacting with the LLM—adding and removing terms from the existing prompt—we were then able to force it into photorealistic mode. It was interesting to observe the LLM’s “reasoning” process as it weighed the pros and cons and ultimately rejected certain options...

Originally Posted by: dsv4600

Anyway what I did learn is that the best prompt is a long prompt, written like it's a scene from a novella, and that's exactly the kind of prompt that works well for Z-Image it seems, so I can see why you followed the LM Studio route to generate more detailed prompts from a "first guess". I have installed LM Studio and the model and will give it a shot.

Well, AI is based on statistical methods. And if you write a lot of text, there’s a greater chance that the right terms will be included :)

Or to put it another way: At some point, you realize that you can remove text without changing the image.

And this is exactly where LLM comes into play again. It distills the text down to the point where generative AI can make the most of it.

And the process is truly interactive. From an image idea that you share with the LLM, through intermediate results that you discuss with the LLM again, to developing additional features, all the way to the finished image.

In the end, it actually feels like you’re working with a capable buddy (you should ask the LLM to stop with the fawning and express itself in a slightly more matter-of-fact way).

But okay, I can understand that communicating with a machine on such a “human” level might seem a bit eerie to some people ;)

The trick is simply not to forget that you’re talking to a machine. The machine is just working for you. And, hey, you can even yell at the LLM :)

Originally Posted by: dsv4600

 All I have to do now is reorganise my SSD to try and get back some of the space all that stuff took up! ;)

If you keep at it, it'll keep growing :) 
I'm now at 1.4 TB on a 2-terabyte M.2 drive. That includes pictures and LLMs, of course.