AI images
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (43)
sorted by:
So any programmers know how long till the whole extra fingers is more or less resolved?
It's pretty much already done, for stable diffusion. Just use a Textual Inversion like Bad-Hands-5 in the negative prompt.
The problem is the big online AI don't have the image to image, masking, and inpainting capability, like Stable Diffusion. With SD you can work with an image until it is good. With bing or who ever else, you only get one shot to get the correct image.
Mostly solved, but you need to be observant. MidJourney and Stable Diffusion allow you to edit photos till you get what you want and then upscale.
Once people start posting more pictures of just one hand.
The fundamental issue is the image AI can't count.
So more 5-fingered hands will just increase the likelihood of a normal hand not actually solve the issue.
LOL the first way I parsed this was "how long until all the top programmers have 'extra fingers' and you can't work as a professional programmer without them?"
If you actually know what you're doing it's a solved issue in many situations. This type of image is a bit tricky because there are so many people in the image and they are further from the viewer than average. If you were to generate this at a larger scale, one section at a time, it probably wouldn't be an issue.
Also, something like this is super easy to fix with inpainting.
Odds are someone was just lazy and grabbed the first image that came out similar to what they were thinking of.