Page 34 of 37

Re: Viridian's AI Experiments

Posted: Wed Nov 29, 2023 8:15 am
by Viridian
These were QSA's originals for the last set.

Re: Viridian's AI Experiments

Posted: Wed Nov 29, 2023 8:16 am
by Viridian
And the set before that.

Re: Viridian's AI Experiments

Posted: Wed Nov 29, 2023 12:16 pm
by Aiko
Impressive progress, but I think the AI is having some difficulty distinguishing mud from cloth - especially in "Spongy Earth v2".

Did you add any prompts to these or was this rendered based on original image + style selection alone?

Re: Viridian's AI Experiments

Posted: Wed Nov 29, 2023 2:00 pm
by Viridian
Aiko wrote:Did you add any prompts to these or was this rendered based on original image + style selection alone?

Yes, I use my set of prompts. The AI won't automatically recognise what it's looking at on its own. There's a fair amount of layering involved to get the right output, but the progress made is in reducing the amount of manual tweaking and redrawing to be virtually none. Without specific prompts, it might mistake mud for cloth, hair for headscarves, metal bikinis for normal bikinis, etc.

In the below outputs, there's no way it would have recognised the setting to be nude beach volleyball and quicksand on its own.

Re: Viridian's AI Experiments

Posted: Wed Nov 29, 2023 9:43 pm
by cerberus
Viridian wrote:These were QSA's originals for the last set.


Ah, thanks. I remember them now. A few more nice conversions as well!

I've been trying a few similar things, some from drawings and artwork (I won't post these as I don't have permission from the artist and in some cases have no idea who the artist is) and a few from my photomaips, with which I've had mixed results (I'll post some of these on my own thread). You'd think the photomanips would be a better starting point, but that doesn't seem to be the case, ControlNet really only uses the supplied image to pose the figures in the output and a photo can be a bit "busy" for that kind of thing. Often simple line art works a lot better as there is less chance of ambiguity.

Re: Viridian's AI Experiments

Posted: Fri Dec 08, 2023 10:01 am
by Viridian
A couple more AI renditions of QSA.

Re: Viridian's AI Experiments

Posted: Fri Dec 08, 2023 10:53 pm
by cerberus
These worked out nicely. I like the pose in the second, probing for the bottom with her foot. She hasn't found it yet and may get a bit of a surprise when she tries to stand up!

Re: Viridian's AI Experiments

Posted: Fri Dec 29, 2023 12:59 am
by DangerGirl
It's really cool seeing QSA's works being brought to AI life.

Re: Viridian's AI Experiments

Posted: Fri Dec 29, 2023 6:58 am
by Viridian
On the note of bringing old art back to life, and in time with Acidtester's comeback, here's some old AT covers.

Re: Viridian's AI Experiments

Posted: Fri Dec 29, 2023 10:21 am
by Jinn
These are awesome. Fantastic to see the infamous redhead come to life.