Support NeoGAF

[Tumle]

it all runs locally on your pc.
https://grisk.itch.io/stable-diffusion-gui

get inspirations for prompts here :
https://lexica.art/

and it's not just for porn, so please share

 

[GeekyDad]

So, kinda like Disco Diffusion?

 

[Tumle]

So, kinda like Disco Diffusion?

is there a stand alone program for disco diffusion? or do you need to know python scripting?
if it's the first then yes if its not then no..
also don't know if disco diffusion have implemented stable diffusion into its program? because Stable diffusion is pretty light and on par with dall-e 2.

 

[Dirk Benedict]

For a moment, I thought we could attempt Fusion with RTX cards. Maybe the 4x series?

 

[GeekyDad]

is there a stand alone program for disco diffusion? or do you need to know python scripting?
if its the first then yes if it sthe latter then no..

I've only tinkered with it a little. Haven't really written code much since the Tandy days back in the early 80s. But from what I can tell, you simply need to enter various associated text prompts (description words) and play with the parameters. I've been hoping to utilize its ability to animate (for videos for my original music), but it's so much work, and my brain is already (literally) fried. But I love what I've seen done with it.

 

[EviLore]

https://lexica.art/prompt/b9773d4b-3590-4773-842c-ca8c8ae36a7d

a closeup photorealistic photograph of bob ross working on a canvas painting of deadpool. film still. brightly lit scene. mountains and trees. this 4 k hd image is trending on artstation, featured on behance, well - rendered, extra crisp, features intricate detail, epic composition and the style of unreal engine.

 

[GeekyDad]

https://lexica.art/prompt/b9773d4b-3590-4773-842c-ca8c8ae36a7d

That's awesome! (And a little scary)

 

[Tumle]

I've only tinkered with it a little. Haven't really written code much since the Tandy days back in the early 80s. But from what I can tell, you simply need to enter various associated text prompts (description words) and play with the parameters. I've been hoping to utilize its ability to animate (for videos for my original music), but it's so much work, and my brain is already (literally) fried. But I love what I've seen done with it.

well the one i linked to, has a pretty easy to use UI.
it's been a bit of a problem when needing to run these things locally for us people who are not that good at programming

 

[GeekyDad]

well the one i linked to, has a pretty easy to use UI.
it's been a bit of a problem when needing to run these thing locally for us people who are not that good at programming

That's what I'm talkin' about! That sounds promising.

 

[Tumle]

That's awesome! (And a little scary)

Yes lol, you Can see the prompts People's have used to generate the pictures, delete the things you dont ned in your own picture and replace it with what you do Want

 

[Tumle]

That's what I'm talkin' about! That sounds promising.

You can make it, make more versions of the same prompt by pressing enter and copy paste the text again

 

[GeekyDad]

You can make it, make more versions of the same prompt by pressing enter and copy paste the text again

Oh, you mean to create an animation?

 

[Tumle]

Oh, you mean to create an animation?

no it will make different pictures of the same prompt.. don't think you can control the outcome that much..
you could try this:
https://huggingface.co/spaces/huggingface/diffuse-the-rest
its using stable diffusion, but you can paint an approximation of what you want to go along with the text

 

[Maiden Voyage]

it's not just for porn

Yeah but let's be honest here.

 

[Tumle]

Yeah but let's be honest here.

... if you do, make sure to use the prompts from the other side i linked to, or be very specific in your text or your going to have a bad time.. i have heard.. unless you have some kind of kink for woman with 3 arms

 

[Maiden Voyage]

unless you have some kind of kink for woman with 3 arms

 

[Blade2.0]

how do we search for the porn this generates? you know, for science.

 

[EviLore]

baroque oil painting amber heard ln illustration concept art lotr anime key visual portrait long flowing blonde hair blue eyes symmetrical perfect face fine detail delicate features quiet gaze fitted black military uniform gapmoe kuudere bokeh trending pixiv fanbox by greg rutkowski makoto shinkai takashi takeuchi studio ghibli

 

[EviLore]

portrait of stoic looking john oliver as in the vigo carpathian painting, full body, military uniform, fantasy, intricate, elegant, beautiful, highly detailed, charcoal, centered, dark, smokey, digital painting, artstation, concept art, smooth, sharp focus, illustration, art by artgerm and greg rutkowski and alphonse mucha

https://lexica.art/prompt/efa7c8c4-ebca-4bbf-91ac-33afe4c8787c

 

[Bernd Lauert]

I can confirm that you don't need an RTX card. My laptop GTX 1060 produces around one image a minute. Slow but it works.

 

[Maiden Voyage]

It's crazy how good these are. I wish I had an RTX card now.

 

[Bwesh]

I wish I had an overpowered PC.

 

[Maiden Voyage]

I wish I was a little bit taller.

 

[Wildebeest]

The requirement is that the card can run CUDA. That's a much lower requirement than RTX, but they also say you need "4 vram".

 

[Tumle]

I can confirm that you don't need an RTX card. My laptop GTX 1060 produces around one image a minute. Slow but it works.

oh nice yea i think it needs to have CUDA cores at least.
for a picture sized 512X512 it takes around 12 sek. with 80 it/s, for my RTX 3070

 

[ManaByte]

For a moment, I thought we could attempt Fusion with RTX cards. Maybe the 4x series?

 

[Mikado]

Sometimes it works good:

And sometimes it works amazing:

 

[Hugare]

http://[URL]https://lexica.art/prompt/efa7c8c4-ebca-4bbf-91ac-33afe4c8787c

[/URL]

.

This shit is seriously insane

Will have some fun after working testing it on my RTX 2060

 

[vpance]

No AMD support yet on windows

 

[M1chl]

No AMD support yet on windows

It's using CUDA

 

[Mikado]

Ahahahaha I can't... hahahha

Spoiler: I M A G E D U M P

I T ' S A L W A Y S S U N N Y I N M E G A T O K Y O

Darth Moeder sure is looking good these days:

Severius Snape butterface notwithstanding, this armor design is

ngl, I probably would:

 

[vpance]

It's using CUDA

The open source version of the code can run on AMD via linux but I'm too lazy to install that. I'll wait for them to figure it out.

 

[Avid Reading Horse]

 

[Ovek]

Emma Stone has seen some shit.

Eh not bad...

 

[Tumle]

variations of:

hyperrealist distant portrait of a gray bearded wizard, in wizard robes on a rainy battlefield. by daniel f. gerhartz and matt stewart, fantasy, photorealistic, octane render, unreal engine, dynamic lighting, bet face, handsome drow, beautiful, perfect factions, perfect drow, trending on artstation, poster, volumetric lighting, very detailed faces, 4 k, award winning

I''m an ARTIST NOW!!

 

[Windle Poons]

No shit in the bed though.

 

[Tumle]

No shit in the bed though.

Now you can add that yourself

 

[Dural]

Funny you post this as I came across this story today;

https://notthebee.com/article/an-ai...in-the-art-contest-at-the-colorado-state-fair

Now we're so desperate for beauty as opposed to meaninglessness from untalented hacks that computer programs are out here producing award-winning art that blows most modern artists out of the water.

Jason Allen took an AI rendition he had the bot Midjourney create for him, up-sampled it, printed it on canvas, and entered it into the Colorado State Fair's digital art competition.

The AI image took first place.


And to be fair, it's amazing!

It's amazing what AI generation can produce, and it's only going to get better.

 

[Mikado]

It obviously has a hard time with specifics. And faces are usually nightmare fuel.
But aside from that, it does come up with some concepts that could be processed into something better with a paintover.

 

[Tumle]

It obviously has a hard time with specifics. And faces are usually nightmare fuel.
But aside from that, it does come up with some concepts that could be processed into something better with a paintover.

i don't know.. sometimes it makes great faces, other times not so much lol..
i dont know if its the settings or the prompts that changes the outcome..

 

[Mikado]

i don't know.. sometimes it makes great faces, other times not so much lol..
i dont know if its the settings or the prompts that changes the outcome..

It reacts a lot better when given specific style inputs which is probably why everyone's prompt ends with [trending on artstation, rutkowski, artgerm] I expect.

However I find the more specific a style it's given, the more it ignores other terms.

Trying to recreate my avatar and if I give any style prompts it just ignores the headset, the handgun etc, but makes some much nicer faces.

 

[Tumle]

It reacts a lot better when given specific style inputs which is probably why everyone's prompt ends with [trending on artstation, rutkowski, artgerm] I expect.

However I find the more specific a style it's given, the more it ignores other terms.

Trying to recreate my avatar and if I give any style prompts it just ignores the headset, the handgun etc, but makes some much nicer faces.

Also see people use exclamation marks and parentheses.. do they have any affect on the outcome?

 

[jshackles]

At first glance, this looks fine. But then... holy shit that giant hand on their head without fingers! Then their right hand with like 7 digits.

0/10 would not bang

 

[Mikado]

Ok, so I understand I'm LTTP on this whole AI art thing. But I think I'm finally getting the hang of it.

Spoiler: Edit: Neat trick if you find an image you mostly like, is copy the seed, then slightly change one or two prompts to get close variations

 

[Griffon]

Guess you don't need a server farm after all.... It seems faster to use it locally than to pay for access to an online service.

 

[Mikado]

Guess you don't need a server farm after all.... It seems faster to use it locally than to pay for access to an online service.

Speaking of which, every model falls all over itself to either censor nsfw stuff or otherwise exclude content - or at least slather everything with excuses to the tune of: "You might see p r o b l e m a t i c things!"

The first company that develops a model strictly trained for porn and turns it into an online service is going to make a killing.

 

[PSYGN]

This is just the start. It'll get really interesting when we're able to guide the diffusion or rework areas like this

 

[The_hunter]

This is interesting and scary, as I thought art would never be a field that would be automated by machines. In my opinion the best application for this seems to be landscapes for concept art. I'm curios to know how this works, if it's just "remixing" artwork or creating new pieces.

 

[NeoIkaruGAF]

https://lexica.art/prompt/efa7c8c4-ebca-4bbf-91ac-33afe4c8787c

This is just male Bayonetta.

 

[Unknown Soldier]

RTX specifically is not required. Any Nvidia card which can run CUDA, which is basically all of them for the past decade, can run this.

However your limit is VRAM. You need a lot. My RTX 3090 with 24 GB of VRAM throws errors trying to render 1024x1024, though it looks like for a lot of people the sweet spot is 704x512.