Shrunken women board

I only now noticed that most of the pictures posted so far are on the safer side, please forgive me for these all being naked.
These were all made in NovelAI and were lucky ones out of MANY "the thing" like monstrosities. The AI doesn't seem to understand the concept at all. I haven't been able to get any interaction with giant hands or any other interaction with a normal sized person. Any time I try to have interactions it always gives me horror. But does sometimes randomly give out oksih pictures. That being said most of these still have problems with anatomy and stuff. Also I've been completely unable to do what others seem to be doing by having the AI enhance an existing picture while mostly still looking like the original. But this is still awesome and I'm very exited for the future of this stuff. (Also the story generator doesn't understand the concept either. It can say stuff like oh she shrank down to 8 inches, but then have the sw do normal sized people things without any problems.)











The previous pictures were all from a prompt only. The next ones are prompt+source sw illustration from this: (It's very VSW so if you're not into that you should probably not click on this link.) https://www.dlsite.com/maniax/work/=/pr ... 12146.html

-

Who shrinks you wrote: ↑

Wed Nov 30, 2022 6:40 pm

Here is a short "tutorial" if you are using the NMKD-GUI (all is trial and error, you need patience..)

A guy is unboxing his 1/6 Scarlet Witch doll.


And now a doll i would buy :

Your getting this down! The doll unpacking is up there with your desktop doll! what about a full image like you did with one of my old ones yesterday more you can post the better. I might try converting the Gal Gadot on the opening title page to my story.. More you can post.... I know we will have to go through our trials & errors.

-

Nahh man you rock at both DS and now AI I'm a little envious Cause I realize I prolly will suck at 1st..

Did I see this right you asked to produce 86 pics???? Do you have path options to where you want to save?

dlsitepom wrote: ↑

Wed Nov 30, 2022 5:24 pm

I only now noticed that most of the pictures posted so far are on the safer side, please forgive me for these all being naked.

Tasteful like you did is fine. These are wonderful!

-

Who shrinks you wrote: ↑

Wed Nov 30, 2022 9:28 pm

bobvan wrote: ↑

Wed Nov 30, 2022 8:24 pm

Nahh man you rock at both DS and now AI I'm a little envious Cause I realize I prolly will suck at 1st..

Did I see this right you asked to produce 86 pics???? Do you have path options to where you want to save?

86 is alot, but it is not like you are rendering 86 pictures in Daz studio.You need just some versions to choose from.
You can use the settings menu to set the outputfolder for your pics.
And now Gal Gadot. I had to cheat a bit, it is 90% Stable diffusion.

This is Bobs original image
I tried it as a source image without a mask, and i got 100% fails to get Stable Diffusion to show me Gal Gadot.

So i used Affinity Photo to cut out (not off) her head and rescaled that to 512x512 pixel (stable diffusion works best with 64*x,64*x resolutions)


Very blurry, but i threw this at stable diffusion, with Gal Gadot as the prompt, and ..


Now there was a little postwork in the editor of your choice (and i did not even try to do my best, this is just to give you an idea ):

Like the closeup Also could help to make a famous person's face look better replicated. I'm going to take the revamped WW pic

Who shrinks you wrote: ↑

Wed Nov 30, 2022 6:40 pm

Here is a short "tutorial" if you are using the NMKD-GUI (all is trial and error, you need patience..)

Nice, thanks a lot for this! I'm getting better but I'm getting crashes when I try the "Face Restoration" feature with the GFPGAN method. CodeFormer seems to work most of the time. Any recommendation on Sampler / Generation steps ?

I'll try to create custom Models and Concepts somewhere down the line... time to make my 3090ti work a bit!

I feel like I'm going down the rabbit hole with this tech lol.

foreverlurk wrote: ↑

Wed Nov 30, 2022 10:44 pm

Who shrinks you wrote: ↑

Wed Nov 30, 2022 6:40 pm

Here is a short "tutorial" if you are using the NMKD-GUI (all is trial and error, you need patience..)

Nice, thanks a lot for this! I'm getting better but I'm getting crashes when I try the "Face Restoration" feature with the GFPGAN method. CodeFormer seems to work most of the time. Any recommendation on Sampler / Generation steps ?

I'll try to create custom Models and Concepts somewhere down the line... time to make my 3090ti work a bit!

I feel like I'm going down the rabbit hole with this tech lol.

Thats a nice GPU

-

I worry that looking at all these twisted mutated figures will hurt my ability to recognize real faces. It would also be nice to know what artists it knows, and what ones it doesn't. Typing Takahashi Rumiko gets me attempts at photorealistic japanese girls, not anything like Ranma or Lum.

Sounds like its not all that simple. Good on Who_ for making progress. I would be curious how long it took him to start to get desirable results.

bobvan wrote: ↑

Wed Nov 30, 2022 10:45 pm

Thats a nice GPU

Felt crazy to drop that much money on a GPU, but it's worth it. Just bummed that NVidia doesn't support DLSS3 on the 3xxx series. With 24 GB I'll be able to train my own models (but that's another story, for now I'm just playing with the default 1.5 model)

bobvan wrote: ↑

Thu Dec 01, 2022 3:07 pm

Sounds like its not all that simple. Good on Who_ for making progress. I would be curious how long it took him to start to get desirable results.

Me too... those past few days playing with Stable Diffusion gave me a lot more respect for AI artists / engineers.

foreverlurk wrote: ↑

Thu Dec 01, 2022 5:43 pm

bobvan wrote: ↑

Wed Nov 30, 2022 10:45 pm

Thats a nice GPU

Felt crazy to drop that much money on a GPU, but it's worth it. Just bummed that NVidia doesn't support DLSS3 on the 3xxx series. With 24 GB I'll be able to train my own models (but that's another story, for now I'm just playing with the default 1.5 model)

bobvan wrote: ↑

Thu Dec 01, 2022 3:07 pm

Sounds like its not all that simple. Good on Who_ for making progress. I would be curious how long it took him to start to get desirable results.

Me too... those past few days playing with Stable Diffusion gave me a lot more respect for AI artists / engineers.

As mentioned I picked up a 3070ti laptop BF sale. Replacing my 4.5 yr old with GTX1070. @ home still using my 2017 i7 10 core 64G workstation with a 2080ti. For rendering its quite adequate. I did D/L the AI program but I think I may want to research & do some reading as it sounds like it can get quite frustrating. Having said that it's like anything once you start getting it..

-

Who shrinks you wrote: ↑

Fri Dec 02, 2022 11:56 pm

Different results interesting..

This is so cool. Love all the new pictures! Amazing!

-

more please! I luv this page!

-

Yeah fair enough. Will be interesting when it evolves.

-

Yeah, that technology is still evolving, after having a fun weekend of messing around with it (not even SW-related), just taking pictures with different prompts, reading and understanding the denoising, samplers, etc. It's really fun.

But I've also come across some of the current limitations, like the 512x512 render and sampling (sampling is actually 64x64, compressed), so anything higher res gets really weird because it can only rendy the first 512 block correctly. There's this "high resolution fix", but it also brings it own problems, like repeated faces (very hard to make a scene with two characters from scratch).

Don't get me wrong, this is pure magic and super fun to mess with, but so far I haven't been able to train it to understand the notion of "size difference". I'm sure it'll be able to, eventually. The inpainting mask is a nice workaround in the meantime. Oh and I've read that they are working on voice sythesis and video tech next! Imagine, being able to produce our own animations...

Edit: forgot about the hands... you know, like in hand-helds, or grabbing a SW, etc... Stable Diffusion gets fingers, limbs, arms, sooo messed up and crank out... **shudders** man-made horrors beyond my comprehension / I've seen things you people wouldn't believe, all that jazz. The horror. Apparently the 2.x releases are better with fingers and limbs, we'll see.