>>3138Messing around with this more because I was bored. Fed it dome random pictures from my camera, wasn't comfortable feeding it any real people but I had some misc pictures I wanted to see if it would recognize.
Observations: Really favors girls. Multiple, when things are ambiguous it would just spawn a girl out of thin air.
Really curious what this was trained on. I've messed around with Stable diffusion before and these results are very different than that program. I'm especially curious about the background it created for the chip bag since there wasn't a lot of information.
Lastly, I was surprised it didn't recognize a single figure image. Boys pictured are more stylized so I guess that makes sense, but I feed it several popular female figures like Madoka and it didn't seem to know them. It did though always recognize that they were figures and create stands so I'm happy there.
Anyway, still have positive thoughts. Not sure what the data collection policy is so I'm not feeding it myself. But I can see how cool this is for people who don't have powerful computers. Bonus points for being easy to use.