tupper now at bsky | @dtupper.comさんのイラストまとめ

WD 1.5 is really good for img2img but do not sleep on NovelAI for nothing-but-prompt generation

Sometimes, even the weird initial face shapes can turn into really cool poses and results if you let them run out to completion. On a whim I chose a really f'd up face shape and it ended up inheriting a sharper style.

replaced with "cowboy bebop" iterate for ~5 generations boosting scale each time i re-set my init image

(fyi, i started with a VRChat screenshot of my avatar)

absolutely wild

Even in e5, WD has GREATLY improved style recognition and transfer. I asked the AI to shift styles to Gainax/Eva/FLCL here and its really apparent

I love AI gen as a concepting tool and an iterative exploration tool (also a fun way to generate new VRC icons)

@ClydeMachine @TUBEVRCLUB @Turels_VR im helping

@Silent0264 I've done the same thing!! It's really impressive what you can do with this method. You can just slightly stylize your screenshot, or even iterate on a character design.

All of these came from VRChat screenshots initially:

@WindmillVR @VRChat

uhhhhhhhhh DALL-E is kinda insane

what is midjourney trying to tell me

i tried REALLY hard but as soon as i zoomed back out after drawing the beard i laughed so hard i definitely accidentally hit alt-f4 and flat-palm striked the "DO NOT SAVE" button haha lol sorry guys