Others Installing Stable Diffusion AI on Win10/11

[Deleted member 1121028]

1. What is Stable Diffusion?

You must be registered to see the links

2. Requirement:
-Nvidia card with 6Go+ VRAM and a good chunk of RAM

At the time I wrote this, install method mentioned

You must be registered to see the links

gave me plenty of errors, but it has been edited last night and seems to works 100% now. Should be recommended method has it include ESRGAN for upscale as well as some editor/mask painting for img2img. Also making possibe with 4Go VRAM with few scripts edit. Once installed

You must be registered to see the links

is a good way to start.

I let the Docker method in spoiler below :

Spoiler

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

You should not have to, as it should be already removed, but in case someone need : How to remove NSFW filter & watermark

Spoiler

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

f. Links:

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

You must be registered to see the links

 

[Microtom]

You must be registered to see the links

 

[Deleted member 1121028]

You must be registered to see the links

is also a good place to start, to get the gist of it.
Wouldn't say no to a Gimp/PS plug-in for serious img2img business.

->text2img, shouldn't upscaled 2 time 4x but it was too late when I saw it.

 

[anne O'nymous]

[some samples]

This starts to be really interesting. Need to be bigger in scale and it could become a good alternative for the backdrop/background.

 

[Deleted member 1121028]

This starts to be really interesting. Need to be bigger in scale and it could become a good alternative for the backdrop/background.

Should be. Not really tested interior yet tho. img2img is quite interesting as you can build your image step by step. That being said problem for serious work is the tedious back & forth between PS & browser (it's easy 50% of the task, 30% being generating image).

->img2img


To include AI on a game, more pernicious problem imho, is to keep artistic consistency/integrity. Not sure what the best strategy. Could also be used for 3D texturing too but ain't touched that yet. Overall it's quite fun I must say.

 

[Microtom]

 

[Doorknob22]

I got it to run. Getting it to render anything even remotely erotic and not nightmare inducing seems like an impossible task.

 

[Microtom]

I got it to run. Getting it to render anything even remotely erotic and not nightmare inducing seems like an impossible task.

If you run it locally you need to change the code to remove the nsfw filters. There are instructions on reddit. The AI can draw boobs, but it doesn't seem to be well trained to draw vaginas. Sometimes the images lack coherence too, but as you can see in my image above it can do a great job.

 

[anne O'nymous]

Should be.

Looks good clearly enough.

Not really tested interior yet tho.

Not sure that it would works as interior. The background is expected to be closer to the characters and despite the incontestable quality, the feeling isn't the same that 2D or even rendered 3D ; it would probably feel a bit odd.
Plus the fact that you would be limited to one angle since I doubt that the AI can reproduce the exact same result but from another point of view. It's less a problem for outdoor scenes, especially if you use the image as distant backdrop, but for indoor ones it's not ideal.

That being said problem for serious work is the tedious back & forth between PS & browser (it's easy 50% of the task, 30% being generating image).

If you take in count the time needed to do a good outdoor render that effectively looks unique (so don't rely too much on generic backdrop), I think it's not too much of a problem.

To include AI on a game, more pernicious problem imho, is to keep artistic consistency/integrity.

I talked about this lately in one of the many "Hey, AI will make the CG of the game really soon" threads.
I doubt this will be possible because a decade or more. The AI need to have some kind of memory in order to reproduce the same models again and again, from different angles and/or positions. It's not impossible to do, but it will probably need to mix AI technology and 3D technology ; the AI not directly generating images, but starting by generating meshes that would then serve as references for the future generated images.
But due to the way AI actually works, it would first need a total rewriting. As they are actually "coded", the would generate the meshes, but using them as reference need a radically different approach.

Not sure what the best strategy. Could also be used for 3D texturing too but ain't touched that yet.

Well, backdrops/background as I said, but texturing is also a really interesting idea. And I take the texturing in the largest way possible ; having finally other wallpaper and posters on the already seen hundreds times living rooms and bedrooms would be refreshing.

Overall it's quite fun I must say.

I don't doubt this a single second... reason why I will avoid it like plague, I already lack of time to do too many things, I'll not add another way to loose some

 

[anne O'nymous]

Getting it to render anything even remotely erotic and not nightmare inducing seems like an impossible task.

Well, the post right before yours don't seem this bad. It's not good, the face being a bit odd, and it's limited nudity. But compared to all the other AI we actually have, there's a giant step that have been made ; Dall-E can't make face at all by example.

But well, as I said, many time now, the problem will still be the constancy of the result on a long run.

 

[Deleted member 1121028]

If you run it locally you need to change the code to remove the nsfw filters. There are instructions on reddit. The AI can draw boobs, but it doesn't seem to be well trained to draw vaginas. Sometimes the images lack coherence too, but as you can see in my image above it can do a great job.

Hlky's fork used in the first post, has no NSFW filter & watermark.
I don't remember for the Docker version tho. I'll add this on the 1st post anyway, thanks.

[...]

For interior it depends what you have in mind imo:
-A VN where each shot has its own PoV
-A more simple one where 1+ sprite(s) in front of a background.

Second seems do-able. First one would need some kind of simulated spherical/cubic lens (would be awesome tho).

For artistic integrity, a RPGMaker prototype should be a good candidate to "see" the scale of weirdness AI can generates (some 2D props creations, "parallax" painting, background/battleground - maybe some monsters). Would need already severe editing regardless.

I don't doubt this a single second... reason why I will avoid it like plague, I already lack of time to do too many things, I'll not add another way to loose some

Yeah 'bout that... I should really go back to other stuff. I got distracted again lmao.

 

[Microtom]

Spoiler

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

 

[txonturiv]

Is there any plan for NSFW SD art thread?

 

[Wizard's Kiss]

Tried playing around with a daz render of mine, then using img2img to translate it into what it would look like if it were a painting. Unfortunately I can only render up to 512x512, so I had to crop the render and then stitch them together, which is quite obvious in the image below.


The prompt was something like "woman, naked, red-haired, big tits, big butt, riding a white horse, on medieval city, photorealistic, highly detailed, dramatic lighting, shimmering colors, by Pre-Raphaelite Brotherhood". Denoise strength was around 0.4 and prompt strength was around 10.

If I had a GPU which could make 1920x1080 stable diffusion AI drawings, I feel like I'd find a lot of places I could use it, but given the difficulty in stitching things and my limit of around 512x512, it's fairly limiting.

 

[anne O'nymous]

[A bunch of NSFW samples]

Yeah, it's definitive, clearly struggle hard with faces. What isn't this surprising since it's the hardest part, but it limit the use of the result.
This said, when the face have a limited importance, or is seen from the side, like in Wizard's Kiss try right above, it works. The mouth is still a bit odd, but it become a small detail noticeable only if you focus on it.

Note that I don't say this as criticism. The fact that it's the only flaw is already amazing by itself.

 

[Deleted member 1121028]

Yeah, it's definitive, clearly struggle hard with faces. What isn't this surprising since it's the hardest part, but it limit the use of the result.
This said, when the face have a limited importance, or is seen from the side, like in Wizard's Kiss try right above, it works. The mouth is still a bit odd, but it become a small detail noticeable only if you focus on it.

Note that I don't say this as criticism. The fact that it's the only flaw is already amazing by itself.

Faces are far from perfect but not that much a problem imho. You can make more or less clean portraiture. Prompt weight/Guidance scale/Sampling method, heavily influence results. As well as height/width (for portrait = use a common portrait ratio). The real struggle from my (small) experience are global body proportion, eyes and hands/feets (by far the trickiest, I think 90% ppl just crop out hands lol).



In another subject, a guy from Polyphony Digital made an expirement with Houdini:

View attachment zzz.mp4












If I understand well (and it's far from certain lol), it's an image (generated by stable diffusion) plane mounted on a camera and distorted using depth estimation (

You must be registered to see the links

) and manual lens guess .



Also Isometric timelapse using

You must be registered to see the links

fork (which I know nothing but

You must be registered to see the links

here for when I have time):

View attachment xxx.mp4

 

[Deleted member 1121028]

Seamless textures Blender plugin:

You must be registered to see the links

Stable Diffusion Photoshop Plugin:

You must be registered to see the links

Stable Diffusion Krita Plugin:

You must be registered to see the links

 

[DoubleLust]

going to be interesting watching this AI evolve

 

[Deleted member 1121028]

going to be interesting watching this AI evolve

Not that good lol.
Check your pen/tablet privilege (StableDiffusion big head), feed the automaton pls o/




Also 3D generated mesh:

You must be registered to see the links

 

[DoubleLust]

researched into it a little more, i was curious what it would take to maintain a consistent look...

You must be registered to see the links

Not sure if I feel like lighting my PC on fire trying to train custom AI