Anonymous at Sat, 27 Aug 2022 22:14:46 UTC No. 915589

>>915588
I've seen this used in texturing 3D stuff, but I don't think this belongs here. I guess it's a closer fit than /ic/, but those related threads you've linked are in boards that fit this a lot better than /3/.
also, it looks like a painting after going through the stable diffusion, but it just looks... worse

Anonymous at Sat, 27 Aug 2022 23:22:55 UTC No. 915602

Not 3DCG

Anonymous at Sat, 27 Aug 2022 23:37:36 UTC No. 915607

>>915588

when i img2img it completely changes my original image or it is a blurry mess. can you give me a screen shot of the settings you recommend?

also, is it possible to use this to turn 3dcg into a realistic photograph with a prompt?

thank you sers and good morning

Anonymous at Sun, 28 Aug 2022 00:20:31 UTC No. 915612

Not /3/.
You can feed 3d into it, but the result is not 3dcg.
The tech is neat and all, but it just doesn't fit the board's theme.
/ic/ would be a better fit. Not to mention it'd probably make them seethe beyond belief.

Anonymous at Sun, 28 Aug 2022 00:29:12 UTC No. 915616

>>915602
Yes

You need to use a lower value in the denoising strength (0.3 to 0.5 max) so it can maintain the form of the imput effect

In the promp you need to add words for realistic render. I.e. hdri, hdri lighting, hdri sky, realistic, ultra realistick, 4k,8k, photograph, studio lighting.

Also add famous photography style such as Terry Richardson, so add "by Terry richardson"

Also try to play with CFG values, higher values the A.I. will try to do the prompt exactly as you wrote it and in lower values the A.I. will get creative

>>915589
>>915602
The usual short sight of the 3d artist, soon we will able to train our own datasets and apply textual inversion to it

Anonymous at Sun, 28 Aug 2022 05:00:09 UTC No. 915635

>>915589
>>915612
>>915602

Anonymous at Sun, 28 Aug 2022 05:59:05 UTC No. 915641

>>915635
Is this supposed to be some kind of rebuttal? You're not proving anything. Can you view that scene at any angle you want? No, you can't. Not 3D.

Anonymous at Sun, 4 Sep 2022 23:16:51 UTC No. 916666

>>915641
>>915588
OP makes a 3D render which looks lacking but then get a better image with AI.
If all you do is doing renders the AI is not that bad.

Hell you can even use the AI image to improve YOUR 3D model, look at the skin painting of the AI, its pretty good in comparasion to the flat colors the original is using, look at the wings, the AI is telling him that they look off light should pass thru them.

The point is that you can use effectively if you think outside the box. swallow your pride Anon, AI is here to stay, use it.

Anonymous at Sun, 4 Sep 2022 23:28:40 UTC No. 916667

>>916666
>6666 quads
oh no, forget what I said. AI is evil and not 3D!

Anonymous at Mon, 5 Sep 2022 00:05:20 UTC No. 916671

>>915635
Can't wait to generate loads of twin porn. Wish this shit was around when I was younger.

Anonymous at Mon, 5 Sep 2022 02:38:29 UTC No. 916689

Could you model/sculpt this?
At least the AI has some understanding of a model sheet, probably much better results are possible with a better and more specific prompt

Anonymous at Mon, 5 Sep 2022 19:37:01 UTC No. 916759

>>916689
>Could you model/sculpt this?
yeah, but why would you?
it doesn't look better than a random sculpt that you can do in 30 minutes.

Anonymous at Tue, 6 Sep 2022 07:46:17 UTC No. 916850

Soon

Anonymous at Tue, 6 Sep 2022 07:52:02 UTC No. 916852

>>915635
Too. OLD.

Anonymous at Tue, 6 Sep 2022 09:08:16 UTC No. 916857

>>916759
>why would you
Well obviously the character looks like trash and you would never actually take the time to model it.
But it's kind of interesting that the AI can create a coherent front and side view in model-sheet style, with reasonable consistency between the different angles.

Anonymous at Tue, 6 Sep 2022 09:46:50 UTC No. 916860

>>916850
what is this? I have never heard of such a method.

Anonymous at Tue, 6 Sep 2022 10:55:45 UTC No. 916863

>>916850
fake and faggot

Anonymous at Tue, 6 Sep 2022 12:45:09 UTC No. 916883

https://reddit.com/r/blender/comments/x75rn7/i_wrote_a_plugin_that_lets_you_use_stable/

!!!!!!!!!

This looks it would actually be godlike for concepting, I wanted for dalle to generate detailed images from my blockouts but it was changing the composition, but this seems to do it properly.

Anonymous at Thu, 8 Sep 2022 17:00:42 UTC No. 917233

>>916850

That's a scam. They outsource the modelling to India and pretend their "AI" did the work.

Anonymous at Thu, 8 Sep 2022 21:09:52 UTC No. 917262

Daz to SD

Anonymous at Thu, 8 Sep 2022 21:21:18 UTC No. 917265

>>917262
Right looks the same, just worse texture detail.

Anonymous at Thu, 8 Sep 2022 21:35:14 UTC No. 917269

>>917265
right is an improvement, look at the collar, tits, ribbon, skirt, and hair
I hate that this technology is going to enable more daz porn VNs

Anonymous at Fri, 9 Sep 2022 03:37:24 UTC No. 917316

>>917262
That's some nice subtle detail. How did you achieve it?

Anonymous at Fri, 9 Sep 2022 03:59:58 UTC No. 917320

Tick tock https://www.youtube.com/watch?v=1Fke6w46tv4

Anonymous at Fri, 9 Sep 2022 15:49:29 UTC No. 917357

This is just another containment thread. I like it.

Anonymous at Fri, 9 Sep 2022 22:13:01 UTC No. 917382

>>915588
>/g/ is LITERALLY taking our jobs

Anonymous at Fri, 9 Sep 2022 22:23:19 UTC No. 917383

>>917382
Don't worry, it's taking their own (non-existent) jobs just the same as your (non-existent) jobs!

Anonymous at Fri, 9 Sep 2022 22:24:24 UTC No. 917385

Man, fuck. That technology seems to be fucking escalating right now. The jumps in progress are fucking insane and to be honest, scary in several ways.

Anonymous at Fri, 9 Sep 2022 22:53:43 UTC No. 917386

>>917385
Zbrush will probably be the first app to implement AI. There really will be a make art button soon enough.

Anonymous at Sat, 10 Sep 2022 00:25:52 UTC No. 917393

>>917383
Speak for yourself I make $25 per render of feet pics.

Anonymous at Sat, 10 Sep 2022 03:20:27 UTC No. 917403

>>917269
>collar
loses the symbol completely, and now gives her a double chin
>tits
becomes amorphous blobs that vaguely resemble tits
>skirt
Yeah this is slightly better
>hair
definitely an improvement, but looks like dried up spaghetti near the ends

Honestly as I said, the 2 are pretty much the same. Apples to apples, there's no noticeable improvement. Both fail and succeed at different points, but enough to even out between the two. Put together it'd be an overall improvement, but on their own neither is really much better than the other.

>I hate that this technology is going to enable more daz porn VNs
You can look at it that way, or you can look at it from another perspective. Less shitty VNs using the exact same art. Some will just plug it in to clean it up, but I'm sure plenty of others will plug it in to get something completely different than a typical shitty DAZ render.

Anonymous at Sat, 10 Sep 2022 06:35:15 UTC No. 917410

>>917403
With enough iterations it can change whatever pick you have :)

Anonymous at Sat, 10 Sep 2022 06:36:25 UTC No. 917411

>>917385
Yeah , textual inversion theory says that sd can be trained even with 5 pictures

Anonymous at Sat, 10 Sep 2022 06:37:34 UTC No. 917412

Can I get this to work with 2GB VRAM?

Anonymous at Sat, 10 Sep 2022 06:42:34 UTC No. 917414

>>917412
I think so, I have seen people with old cards getting it to work there are some command lines for low vram, but the bigger the better for this stuff

Anonymous at Sat, 10 Sep 2022 13:58:27 UTC No. 917427

This is from some Anon over in the /g/ thread.
Fucking hell shit's getting real

Anonymous at Sat, 10 Sep 2022 14:01:17 UTC No. 917428

DAZ3D is in deep trouble now

Anonymous at Sat, 10 Sep 2022 14:04:14 UTC No. 917429

Anonymous at Sun, 11 Sep 2022 04:50:24 UTC No. 917486

>>917412
There's already a patch for SD to work on lower end rigs

Anonymous at Sun, 11 Sep 2022 05:53:18 UTC No. 917489

>>917486
Link?

Anonymous at Sun, 11 Sep 2022 13:20:43 UTC No. 917517

>>917428
If it can do vaginas then hentai is dead.

Anonymous at Mon, 12 Sep 2022 02:20:21 UTC No. 917613

>>917517
Maybe I'm just bad at throwing prompts at it, but I've not once got a vagina from SD that doesn't look like gobbledygook.
It can do boobs alright though.

Anonymous at Wed, 14 Sep 2022 16:24:21 UTC No. 918038

>>917613
vaginas aren't really in it's dataset. It's probably possible to train it on like 50,000 images of different vaginas for effect, but you'd need to find someone who both has like 60GB VRAM and is willing to run it for a week straight

Anonymous at Thu, 15 Sep 2022 01:45:19 UTC No. 918098

>>917262
wait i'm also using stable difusion and img2img an it redo all the model without traces of the original...
what i'm doing wrong?
I just want the AI for prototyping my own models but it not working ahhhhhhhh.

Anonymous at Thu, 15 Sep 2022 02:00:16 UTC No. 918102

>>915588
im such a brainlet i dont understand how to do these rentry installs
why so complicated cant someone just make an exectuable?

>t.unemployed archvis neet

Anonymous at Thu, 15 Sep 2022 02:02:31 UTC No. 918103

>>917269
I see consitincy being a pretty big issue for porn VN, no?
Yes, still look good like that sailor moon one, but if you have a character constantly getting redone wont it lead to odd discrepancies, or can the AI be told to keep certain aspects similar to resent img2img inputs?

Anonymous at Thu, 15 Sep 2022 02:12:35 UTC No. 918104

>>918038
just a matter of time then, im sure some coomer is already on it

Anonymous at Thu, 15 Sep 2022 03:19:19 UTC No. 918106

>>918104
I dunno, most of the people training seem to be laser focused on anime, furry, and drawn stuff from what I've seen.

Anonymous at Thu, 15 Sep 2022 08:01:23 UTC No. 918121

this ai art actually made me seriously depressed. it's going to take awhile before i come to terms with artists being raped like this
i'm even trying to learn to become a decent writer, but the AI will (and most already has) leave me in the dust

Anonymous at Thu, 15 Sep 2022 09:06:25 UTC No. 918126

>>915588
It's not perfect, but it does look like a good way to come up with new references to base future projects on.

Anonymous at Tue, 20 Sep 2022 05:38:55 UTC No. 918863

>>918121
>but the AI will (and most already has) leave me in the dust
Not for a while, but eventually yes AI will leave humanity in the dust.

Anonymous at Tue, 20 Sep 2022 05:46:15 UTC No. 918867

>>917262
Holy fucking shit first /ic/ and now /3/ is also dead.

Anonymous at Tue, 20 Sep 2022 05:47:34 UTC No. 918868

So should I get the normal trained set or the danbooru one? It's going to be censored either way right?

Anonymous at Tue, 20 Sep 2022 06:37:25 UTC No. 918872

>>918867
/3/ is even slower than /ic/ these fa/g/s honestly are worse then the hugboxes on fuckin tumblr

Anonymous at Tue, 20 Sep 2022 07:07:24 UTC No. 918873

>>917262
How did you get it to do subtle detail while leaving most of it normal? Did you infill it?

Anonymous at Wed, 21 Sep 2022 06:39:27 UTC No. 918996

>>918121
Look at it differently (cope), by thinking how you can make something that has never been done before. You can make hundreds of amazing images over night. What can you use this for?

Look at this

https://www.youtube.com/watch?v=XWATqtM5Tdw

Imagine having to draw every frame manually, it would have taken ages, yet this guy just rendered the entire thing in a few hours. There is room for figuring out something that has never been done before using this tec. Experiment with it.

Anonymous at Wed, 21 Sep 2022 11:17:13 UTC No. 919025

>>917428
explain to an IA how to compose that

Anonymous at Wed, 21 Sep 2022 11:23:55 UTC No. 919026

preddy neato

Anonymous at Wed, 21 Sep 2022 11:24:56 UTC No. 919027

>>919026
https://github.com/FrozenBurning/Text2Light

Anonymous at Wed, 21 Sep 2022 11:40:50 UTC No. 919028

>>919025
don't bother, AI-bros will not understand.
Making a headshot look pretty is literally miles better than doing a composition with multple characters the way you want to.
They don't realize that this AI will make our work better and they cope on how their rng god is just another tool for us in the toolbox.

Anonymous at Wed, 21 Sep 2022 11:48:54 UTC No. 919029

found this seed traveling script, but I can't get it to work properly. It generates stuff that looks like it would "fade" into the next seed, but then it just jumps to the next image.
In the video that does happen as well, but kind of smoother? I dunno.
Any help?
https://github.com/yownas/seed_travel

Anonymous at Thu, 22 Sep 2022 19:18:47 UTC No. 919241

>>919025

Anonymous at Sun, 25 Sep 2022 04:56:12 UTC No. 919555

>>919025
Don't need to, you have the AI generate each character separately then just Photobash them together.

Anonymous at Thu, 29 Sep 2022 13:44:18 UTC No. 920153

>>919241
Soul.

>>919025
Soulless.

Anonymous at Tue, 11 Oct 2022 07:05:45 UTC No. 921673

>>920153
You honestly think there is soul is some shitty 3d models dragged onto screen with some pose presets made for coomers? I promise you if you don't tell the coomer its AI generated he couldn't tell the difference. This shit is the lowest tier of "creativity"

Anonymous at Wed, 12 Oct 2022 20:09:21 UTC No. 921802

>>921673
>reading comprehension

Anonymous at Wed, 19 Oct 2022 12:41:19 UTC No. 922456

2img

Anonymous at Wed, 19 Oct 2022 13:23:21 UTC No. 922458

AI is so fucking boring

Anonymous at Fri, 21 Oct 2022 09:43:02 UTC No. 922709

>>915588
I'm using the SD colab:
https://colab.research.google.com/drive/1jUwJ0owjigpG-9m6AI_wEStwimisUE17?usp=sharing

any way to add negative prompts?