Shrunken women board

Little Sally wrote: ↑

Fri Jan 19, 2024 10:18 pm

The one I mentioned that I first tried and liked, had the option where you could upload a pic (including work of your own) and AI would do interpretations of it. Or reimagining it as they like to term it.

Oh so you could upload a picture, and prompt something based on the that picture? That's usually done through img2img w/ a "Denoise level" - AIs don't draw the same as humans do, they learn to remove chaos (ie. Gaussian Noise) rather than by adding, so the Denoiser's job in this case is to adhere to the original pic up to a certain threshold (0 == "don't touch anything", 1 == "free for all").

Maybe those details are hidden in the online version but I'm pretty sure they're Stable Diffusion models. Doesn't narrow it down because there are dozens online who offer that service.

Pacific1982 wrote: ↑

Fri Jan 19, 2024 11:25 pm

Should also mention I really love the image of the girl sitting on the piano keyboard. That is one of my preferred subject matters. Although I’d also prefer her to be nude.

It's a favourite of mine, too. Goes way back. I guess I could generate some nudes, I'll see what I can do.

foreverlurk wrote: ↑

Sat Jan 20, 2024 5:08 am

Oh so you could upload a picture, and prompt something based on the that picture? That's usually done through img2img w/ a "Denoise level" - AIs don't draw the same as humans do, they learn to remove chaos (ie. Gaussian Noise) rather than by adding, so the Denoiser's job in this case is to adhere to the original pic up to a certain threshold (0 == "don't touch anything", 1 == "free for all").

No, it had the usual text prompt option as the main thing, then in addition the option to upload an image for reimagination (or re-interpreting) as a separate thing. So two ways of obtaining an image, but neither complimenting the other If that makes sense. I don't think there was a way of influencing what AI did with an uploaded image (at least I didn't notice that), it was just a case letting it do its thing and seeing the result.

foreverlurk wrote: ↑

Sat Jan 20, 2024 5:08 am

Is it possible you were editing the generated pics in Microsoft Designer? It's yet another way to access DALL-E:
https://designer.microsoft.com/image-creator

Thanks for linking to that one which is great. It isn't the one with the pic upload option but still good fun!
I amused myself getting some more octopus related stuff from it (no it isn't a new fetish for me, just a passing whim lol!)

I think my prompt was "a smiling octopus dancing with a woman in a dress in a realistic setting"

Little Sally wrote: ↑

Sat Jan 20, 2024 9:43 am

I think my prompt was "a smiling octopus dancing with a woman in a dress in a realistic setting"

These are so funny an creative, Sally. More I say, haha!

Pacific1982 wrote: ↑

Fri Jan 19, 2024 11:25 pm

Should also mention I really love the image of the girl sitting on the piano keyboard. That is one of my preferred subject matters. Although I’d also prefer her to be nude.

Here's one I liked (I have a few more I'm still tweaking) :

Beautiful work in all the images, I really like the settings and the interaction!

blueman wrote: ↑

Mon Jan 22, 2024 11:19 am

Beautiful work in all the images, I really like the settings and the interaction!

Thanks! I love interaction between tiny hands and normal sized fingers...

Beautiful!

A few more I recenty managed to create, in spite of the recent nerfs and stricter filtering in DALL-E.

Amazing job, a whole new World awaits SW arts

Tried the new Google AI called Gemini yesterday - wasn't impressed, even more censored than DALL-E, and it did lecture me on the ethics of shrinking, lulz.

Pacific1982 wrote: ↑

Fri Jan 19, 2024 11:25 pm

Should also mention I really love the image of the girl sitting on the piano keyboard. That is one of my preferred subject matters. Although I’d also prefer her to be nude.

How about this little lady? Wearing a cowboy hat and a guitar.

Some more AI content, those were all generated before the Taylor Swift deepfakes drama. I don't know if that will ever die down, but DALL-E has not been the same ever since.

Perks of the Job


Bad Idea


Not Another Word


Size Remote

Another sample from my latest update, playing around with a new model and getting good results I think. More on DA.

foreverlurk wrote: ↑

Wed Mar 20, 2024 7:06 pm

Another sample from my latest update, playing around with a new model and getting good results I think. More on DA.

Gonna have to agree with you. Very nice. Really like the last one for some reason.

DocRick wrote: ↑

Wed Mar 20, 2024 10:13 pm

Gonna have to agree with you. Very nice. Really like the last one for some reason.

Yeah I was sleeping on some of the newer tech for way too long. The next version of generative AIs will be scary good I think.

What program are you using now?? Chat gpt and bing image creator has been nerfed:(

Zeshrinkar wrote: ↑

Thu Mar 21, 2024 2:19 am

What program are you using now?? Chat gpt and bing image creator has been nerfed:(

Yeah, "nerfed" is an understatement. I still use ChatGPT to get the base prompt. I need to be way more creative with the prompting nowadays, and I still haven't found workarounds that are 100% foolproof (like those that worked before TaylorSwiftGate).

What changed is my second step in Stable Diffusion. DALL-E is purposely making images less photorealistic (since November), so I used to do img2img through SD version 1.5 and with RealisticVision to get some photo realism back. Now I do the same but with SDXL and JuggernautXL which is what you're seeing in those latest images. All of this is done locally, for user interface I either use ComfyUI with a custom workflow, or Fooocus when I just need a quick touchup (they have a really good custom inpainting/outpainging model).

BTW the next version of Stable Diffusion is looking very promising, maybe it'll finally have good enough prompt understanding for sizey content.

Hi Foreverlurk,

Would you mind sharing the prompt or words to miniaturize? I'm beating my head on the wall and can't seem to make tiny humans unless I start with a 1child and try to age it.

Been messing around with AI myself, testing my program of choice's limitations to see what it can do. Mine has a NSFW filter I usually have to get creative to get around.

djbtyt wrote: ↑

Tue Apr 09, 2024 7:56 pm

Would you mind sharing the prompt or words to miniaturize? I'm beating my head on the wall and can't seem to make tiny humans unless I start with a 1child and try to age it.

Sorry I missed your post somehow? Can you show me what you are trying, and on which AI? It's all very contextual, but I can help if you show me examples.

dubiouskey wrote: ↑

Thu May 02, 2024 10:09 pm

Mine has a NSFW filter I usually have to get creative to get around.

Oh, what are you using? Is it a local setup or online? I guess it's online since you're talking about filters?

Sample of my most recent work, there’s more on DA if you are interested. Still waiting for Stable Diffusion 3 and praying to the AI gods that they don't go under before releasing the weights.

Enthomology


Pleasure, Little Treasure


Same Difference


Healthcare Reduction


Lost and Found

foreverlurk wrote: ↑

Thu May 16, 2024 9:18 pm

Sample of my most recent work, there’s more on DA if you are interested. Still waiting for Stable Diffusion 3 and praying to the AI gods that they don't go under before releasing the weights.

Not a fan of AI imagery per se, but these represent the use of some intelligent imagination. So, nicely done sir! (and I like the last one best of all).

They are for instance much more preferable than some of the "AI" examples posted recently, using celebrity "clones" in explicit porny poses. I think users should tread carefully of following that particular path.

Little Sally wrote: ↑

Thu May 16, 2024 11:13 pm

foreverlurk wrote: ↑

Thu May 16, 2024 9:18 pm

Sample of my most recent work, there’s more on DA if you are interested. Still waiting for Stable Diffusion 3 and praying to the AI gods that they don't go under before releasing the weights.

Not a fan of AI imagery per se, but these represent the use of some intelligent imagination. So, nicely done sir! (and I like the last one best of all).

They are for instance much more preferable than some of the "AI" examples posted recently, using celebrity "clones" in explicit porny poses. I think users should tread carefully of following that particular path.

Too many AI art posters post a bunch of slightly different pictures off the same prompt, which makes it all look very samey. While I do have issues with AI art in general, at least these show some effort put into the prompts and look like pictures worth doing.

Little Sally wrote: ↑

Thu May 16, 2024 11:13 pm

Not a fan of AI imagery per se, but these represent the use of some intelligent imagination. So, nicely done sir! (and I like the last one best of all).

Thank so you much for your kind words. I'm really proud of the last pic, glad you liked it. I strive to bring something original, simply by tapping into to my own fantasies. I'm also not using celebrities, personally I'm just not comfortable with using real persons without their consent, famous or not. But if a shrunken lady would ask me nicely to create such pictures of her resemblance, then I would certainly oblige.

LordAsteroth wrote: ↑

Fri May 17, 2024 1:54 am

Too many AI art posters post a bunch of slightly different pictures off the same prompt, which makes it all look very samey. While I do have issues with AI art in general, at least these show some effort put into the prompts and look like pictures worth doing.

Thanks, and I understand perfectly reservations about the use of generative art. My stance has always been to be upfront about my usage of these techs. I'm not seeking to profit from this either. I'd rather have one or two good pictures and put lots of hours into tweaking them, than flooding my gallery with hundreds of low effort lookalikes.

Those last two.
Wow...

I agree with the whole not-using-celebraties thing. For me, one of the appeals of AI art is that they aren't photos of real people, so you don't have to see behind the image and worry why the originals were taken.