HTML Demon Gods [v0.5] [Panonon]
It would change the art style again, but if it can save me editing time with more reliable fingers / etc, then that would be really nice.
I enjoy making the art but there are many scenes without any art because of how long it takes me to edit them.
Especially the 2+ person lewd scenes.
Ah yeah, I mixed up the names in the code iirc a while back.
Thought I fixed that but I guess not.
Thanks.
What's the current content of pseudinferis?
Necessary dead or another option instead of kill?
Since there's evil karma, it just make wonder that's all.
Necessary dead or another option instead of kill?
Since there's evil karma, it just make wonder that's all.
- Mar 7, 2019
- 37
- 22
Funny enough... I'm making some gangbang scenes with that character I posted earlier right now. It also has a handy-dandy tool called "in-paint" that lets you mask an area for the AI to "fix," I've used it a bit so far to fix hands, extra fingers, extra... insertion points LOL.
Plus another tool where you can "vibe" one image into another to keep styles consistent. The tags "ai generated, ai assisted, midjourney" or Stable Diffusion/3D/Blender (medium) might be able to keep the style consistent with what you have already. There's a ton of stuff in their
You must be registered to see the links
.And they're working on a version specifically for furry stuff for when you need it.
Goddamn I sound like a salesperson but I've been following NovelAI since it first started and the devs are just awesome. Can't recommend it enough. (Plus it's a two-in-one subscription with ai-assisted writing)
Some more examples:
You don't have permission to view the spoiler content.
Log in or register now.
Pseud dies no matter what you do.
But you can kill him without fighting him based on certain decisions.
Tbh it would actually be a bit harder to isnta-kill him if you're doing an evil route. Still possible, though.
Those images you provided are much better than I can get with Stable Diffusion, even if I use controlnet.
I usually end up having to manually edit them myself.
What were your prompts? And how much did you need to fix in inpaint?
I use inpaint a lot in SD, but it's tricky. Can't really fix hands in inpaint, even if I use controlnet. Still gotta do those manually.
Same with eyes, and certain other things.
- Mar 7, 2019
- 37
- 22
Using inpaint can have varying results, but I found that I had less issues on hands with NSFW images than SFW images, funnily enough. In my previous post, the top image has no inpaint usage at all. For the bottom image, I used inpaint to remove an extra asshole, but that's really it. Usually I'd end up using inpaint 2 or 3 times, swapping back to other images if the results became too distorted. It's best at removing extra fingers, but sometimes messing with very small areas works, too.
Here's a couple images side-by-side utilizing inpaint to fix the hands and one Enhance Magnitude 1 at x1.5 res, it's not perfect but it would only require minor editing to fix:
Using the Enhance function at magnitude 1 usually fixes most eye problems I've had.
For my previous post, the prompts where as follows:
Here's the gangbang ones I was working on before, with prompts
I also did some more with a piledrive position, just used "upside down" tag near the top.
Here's a couple images side-by-side utilizing inpaint to fix the hands and one Enhance Magnitude 1 at x1.5 res, it's not perfect but it would only require minor editing to fix:
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
Using the Enhance function at magnitude 1 usually fixes most eye problems I've had.
For my previous post, the prompts where as follows:
You don't have permission to view the spoiler content.
Log in or register now.
Here's the gangbang ones I was working on before, with prompts
You don't have permission to view the spoiler content.
Log in or register now.
I also did some more with a piledrive position, just used "upside down" tag near the top.
Last edited:
Yeah, Stable Diffusion tends to distort fingers and hands even further if you try to fix them with inpaint, so I've always just done it by hand.
Here's an example of what Stable Diffusion (with my own custom model) spit out with your prompt (and I matched your settings as close as possible.
Stable diffusion has a few extra settings that I left in a safe middle ground though).
You don't have permission to view the spoiler content.
Log in or register now.
These were the best two, everything else was absolutely useless.
But these are a great example of how much manual editing I have to do with any image stable diffusion generates for lewd scenes.
That's why it takes me so much time and effort to complete a single lewd image. Takes me a couple of days usually to finish editing characters into a coherent scene.
Edit:
Oh, Anlatan is NovelAI.
I didn't realize they had improved that much. I don't like their writing tool, and originally stable diffusion was a lot more customizable than novelai so I only briefly tried it (a couple original images in this game that were removed were originally from novelai).
But if they've improved this much, I might try them again.
Edit 2:
Looks like I still had some tokens left over from a couple years ago on novelai (cool of them not to expire my old tokens).
Tried your prompt and I am honestly impressed.
You don't have permission to view the spoiler content.
Log in or register now.
The only downside is it still hides the mens faces but the amount of editing I'd need to do in photoshop is drastically decreased.
I might resub with them again just for the image generator.
Downside is that it would unfortunately change the art style again.
Last edited:
- Apr 14, 2021
- 5,070
- 12,901
Of course I agree, but I was hoping to see them eventually switch to an LLM style image generator like Dall-E.You should try everything again once in a while, things rarely stand still after all.
That said, it's impressive enough that I'm willing to pay a sub to make my life a little bit easier to make some images for the game.
I should be able to finish chapter 3 this month then next update I intend to write a linear opening to the next chapter and focus on adding a lot of missing lewd images. I think I'll give this a shot for the images.
- Mar 7, 2019
- 37
- 22
Glad you're impressed! (holy shit those stable diffusion shots are ugly af. It really doesn't like multiple characters). I was working on prompts for getting multiple characters with faces, with some moderate success. I think hiding male faces might just be an issue of the art it's pulling from. A lot of hentai and doujinshis just don't put a ton of detail into the guys' faces, especially with multiple partners.
Regular SFW prompts seem to be doing pretty OK with multiple characters, but it's a mixed bag. Maybe I'd need to generalize the clothing a bit more or use inpaint to literally "place" a character in a spot using a base image. So it'd be like "Generate this image with this character" then use that image as a base image, generate, inpaint where you want the second character to be, generate the new character with a different prompt.
I have not idea if this would work. Haven't tried it yet. OFC if a character is nude you'd need a lot less tags to generate the character properly. More than half the memory on my prompts are clothes.
About the artstyle change, I would be uncomfortable with that, as well. It all depends on how much time you save and whether or not you would go back and change the old art or try to get Anlatan to match your old art style as much as possible and forgo going backwards. Personally, from my own development experience, I would just continue onwards and go back near the end as touch-up. Going backwards can be demoralizing, but being close to done and happy and then going back to make things better would be less upsetting for me, personally.
Regular SFW prompts seem to be doing pretty OK with multiple characters, but it's a mixed bag. Maybe I'd need to generalize the clothing a bit more or use inpaint to literally "place" a character in a spot using a base image. So it'd be like "Generate this image with this character" then use that image as a base image, generate, inpaint where you want the second character to be, generate the new character with a different prompt.
I have not idea if this would work. Haven't tried it yet. OFC if a character is nude you'd need a lot less tags to generate the character properly. More than half the memory on my prompts are clothes.
About the artstyle change, I would be uncomfortable with that, as well. It all depends on how much time you save and whether or not you would go back and change the old art or try to get Anlatan to match your old art style as much as possible and forgo going backwards. Personally, from my own development experience, I would just continue onwards and go back near the end as touch-up. Going backwards can be demoralizing, but being close to done and happy and then going back to make things better would be less upsetting for me, personally.
Last edited:
Yeah, I haven't played a lot of japanese VN's but from the few I've played the scenes were generally like that.
If I were to use controlnet to help guide the scenes I have much better luck but I'm usually still redrawing legs / boobs / hands / etc, and sometimes merging pieces of seperate images together to create a single coherent image.
A lot of times I'll pre-draw scenes, then use stable diffusion to try and recreate those scenes. With some success here and there, but usually requires me to merge multiple scenes together.
Stable Diffusion is best at portraits. I still have to fix pieces of those often-times, but it's a lot less work and the reason I can easily add new characters rather than scenes.
Actually looks like the art-style change isn't that severe. Old NovelAI would never have worked, but the current AI model could actually merge pretty well with the current games art style while also steering it back to an anime style that I was accidentally falling out of.
It's a shame novelai still doesn't have controlnet features, but it's already a big improvement.
It took me a long time to get around to this, But I have finally fixed it so they expand on click instead of hover (in the unreleased version 0.52).
It's been on my to-do list since forever ago, but since I was also fixing text size / color, I took the time to fix this too.