Home Artists Posts Import Register

Content

I'm unsure how far to go with this tool, as I myself have been frustrated with the flood of same-y, weird looking portraits on deviant art and other sites lately.  I did this piece with img2img, based on this, but I cranked the denoising slider way up to a result in a different style.  

It took several rounds of in-painting and manual tweaking.  I rerolled her hair several times before discovering the magic set of words to make it suitably poofy (it's "farah fawcett").  At some point her collar disappeared so I redrew it and refiltered to make it look cool.  I Eventually gave up on the ring on the front, since I just couldn't get it to understand the structure.  Still, I'm not sure this counts as "work" and it feels a little weird to post something so far outside of my normal style and beyond my painting abilities.

I've experimented some and the image generation doesn't seem to be consistent enough to use for animation - or even a comic, really.  It might be possible with dreambooth training but I don't have enough vram for that at this time.  I'll be building a new computer soon to pursue this.

In the future, I think editorial will be more important than ever since you (or anyone) will be able to choose from thousands of different looks and styles for everything.  With that in mind, I'm gonna try to make sure any AI picture I post is as free from weird flaws as I can make it.  (Though I might post some bloopers because they can be funny.)  I got a little excited with the initial batch, but I think they could be improved a lot.  

Files

SD 02468

Comments

Ricky Zimmermann

your original Alison has more personality …

Ellarihan

It's just a tool, yes, but we can't even foresee all its possibilities. The time has come for great changes in the arts. I'm glad you're experimenting with neural networks. A lot of other artists hate them.