Why does nobody talk about how AI image generators handle hands better now

I tried Midjourney v6 last week to make a portrait for a presentation and was ready to laugh at the messed up fingers. But every single hand came out with five fingers, all in the right spots, no weird melting. On the other hand, the background objects like coffee cups and chairs still look like they were drawn by a kid who just learned what shapes are. Is it really that much harder to train on backgrounds versus body parts, or did they just prioritize fixing the creepy hand thing because everyone complained? Has anyone else noticed this split in quality or am I just getting lucky with my prompts?

3 comments

3 Comments

kaid5920d ago

Yeah, totally get what you mean. It's like they fixed the most obvious freakout area first. Hands were such a meme for so long it probably hurt their reputation more than bad chairs. Makes sense they'd put all their focus there. People notice hands instantly, a messed up coffee cup you might just scroll past. But the background stuff is still rough. I made a picture of someone holding a book and the book looked like a weird sponge. You aren't getting lucky. The hand thing is just way better now. The rest of the scene is still playing catch up.

alexc9320d ago

Dude I know what you mean, @kaid59. I tried generating a person holding a cup of coffee the other day and the cup looked like it was made out of melted playdough. But the hands on the same image were totally fine, five fingers and everything. It's like they patched one part of the car engine but left the tires flat.

grace_bailey20d agoTop Commenter

Try adding some specific lighting terms to your prompts, like "soft window light" or "studio lighting." It forces the AI to pay more attention to the whole scene instead of just the main subject. Also, sometimes adding "product photography" or "prop" before the object helps, like "prop book on a wooden table" instead of just "book.