[X] My Assistant

So I still don't understand how to use programs to create images as requested in the first topic . Can anyone help with this?

[www.reddit.com] https://www.reddit.com/r/StableDiffusion/co...ion_stepbystep/

And something like [civitai.com] https://civitai.com/ for the models.

There's probably other software now but stable diffusion used to be the main one.

Thank you very much, I'll study it. And there are no such things in Russian?

I'm not sure, you can try asking in the Russian thread, on the off chance someone there knows https://forums.e-hentai.org/index.php?showtopic=193300

I'm experimenting with using ComfyUI inpainting to decensor. The results are okay-ish and generally could use a lot of improvement. I wanted to ask here to get some info from people who know more than I do about stable diffusion (I'm a complete noob).

Here is an example of what I'm talking about (cropped):
(IMG:[files.catbox.moe] https://files.catbox.moe/akmoj3.png)

I have questions about how I could improve my results.

• Should I ditch the standard 512-inpainting-ema and sd-v1-5-inpainting models? Which models should I use?
• Is it possible to get SD to match the artstyle of the artist for more complex decensors and modifying content?
• I can't get SD to generate NSFW stuff with my current settings, especially for mosaic decensors. It'll generate anything except for anime dick and pussy, even if I specify it in the text prompt. How do I fix that?
• What text prompt should I use for bar decensoring?
• What sampler should I use?
• SD changes pixels that are outside of the mask area. How do I stop this from happening?
• I am using the default workflow that appears when you select ComfyUI inpainting. How can I modify the workflow to get better results?

This post has been edited by Zukuyo: Jul 13 2025, 19:07

I want to know if VRAM is lack, did computer will use Unified Memory for running automatic or I has to setup?

I know it's slower than local VRAM but It' better than didn't work at all.


And I hear that RTX5090 and 5080 didn't use unified memory, is it that real?
if it real then can I setup to use it or can't do it at all.

One more question did CPU really important to do the AI Generated? if it does how to choose it.

I want to build a new computer for gta6, and I interest AI Generated so now, I look at 5080s that have a rumor it'll come with VRAM 24Gb.

BUT if you have a suggest I'll be thankful to you, THANK YOU (IMG:[invalid] style_emoticons/default/smile.gif)

I'm experimenting with using ComfyUI inpainting to decensor. The results are okay-ish and generally could use a lot of improvement. I wanted to ask here to get some info from people who know more than I do about stable diffusion (I'm a complete noob).

Here is an example of what I'm talking about (cropped):
(IMG:[files.catbox.moe] https://files.catbox.moe/akmoj3.png)

I have questions about how I could improve my results.

• Should I ditch the standard 512-inpainting-ema and sd-v1-5-inpainting models? Which models should I use?
• Is it possible to get SD to match the artstyle of the artist for more complex decensors and modifying content?
• I can't get SD to generate NSFW stuff with my current settings, especially for mosaic decensors. It'll generate anything except for anime dick and pussy, even if I specify it in the text prompt. How do I fix that?
• What text prompt should I use for bar decensoring?
• What sampler should I use?
• SD changes pixels that are outside of the mask area. How do I stop this from happening?
• I am using the default workflow that appears when you select ComfyUI inpainting. How can I modify the workflow to get better results?

The AI decensor tools lack the following: (after trying several tools):
1. You don't ditch the standard, they are as of now, are the best decensor model.
2. Complex decensors and modifying content are still artists' domains. (at least as of now, August 2025).
3. Mosaic, pixelated, full censorship are still the bane of the decensor tool. Only the small black/white bar is possible currently.
4. Just use your photoshop for bar decensoring, it's easier, not making you more frustrated.
5. You can't do anything to stop this from happening. I tried almost all ways until now, they usually 'change' a bit of outside of the mask area.
6. I don't know. It's an established default workflow.

Here are the full homeworks for AI:

-> It can't draw clitoris/clitoral hood well enough complementing the artist, or when it does, it doesn't connect well with others, and end up just glossing over.
The example pics show it. It only draws labia/vagina. (or penis)
-> It can't draw female urethra/pee hole well, usually it glosses over the details or end up with gibberish.
-> It can't draw (a)symmetrical line and muscle around the vaginal wall or inside just yet (it became strange lines), following realistic pic.

I want to know if VRAM is lack, did computer will use Unified Memory for running automatic or I has to setup?

I know it's slower than local VRAM but It' better than didn't work at all.
And I hear that RTX5090 and 5080 didn't use unified memory, is it that real?
if it real then can I setup to use it or can't do it at all.

One more question did CPU really important to do the AI Generated? if it does how to choose it.

I want to build a new computer for gta6, and I interest AI Generated so now, I look at 5080s that have a rumor it'll come with VRAM 24Gb.

BUT if you have a suggest I'll be thankful to you, THANK YOU (IMG:[invalid] style_emoticons/default/smile.gif)

1. RTX 4090 or more (especially the 32 GB VRAM version) does the job well. or even better than the 5080
2.Total VRAM in graphic card is more important than your actual RAM for AI-related works. If you can combine several graphic cards (RAID), it's even better.


This post has been edited by Frostbite: Aug 28 2025, 14:58

Original title: AI engine for Manga generation

Hi - I am using Chat-GPT to write my own story (it’s fun), and have an inspiration to turn the story to a manga.

Anyone knows what AI platform enables me to do that?

Thanks!

This post has been edited by Shank: Aug 29 2025, 20:49

Can you select Lora models from Stable Diffusion webui's drop down bar? I can't get SD to read any of the models I downloaded from CivitAI.

There should be a refresh button, if you've added them without restarting it, for them to display.

You should also make sure they are in the lora folder, not the folder you put the checkpoints/models

I always make sure I close out SD and end any Python processes running before trying to reload it. I am going to assume my system is stuck in dependency hell with Py packages.