Yeah, I got something similar, but this is no good. The variations are too subtle. To me: one person can be an animal, another can be a simple square (but that is the only square in the set), another can be a single letter. Big themes like professions, nationality, memorable colors, actions

The Brazilian black rabbit astronaut riding a blue bike would be an awesome image.

It might be better to turn the key into words first and then ask the AI to generate layers for those words.

Reply to this note

Please Login to reply.

Discussion

Don't lose signal by giving it a bag of words, design a system with the desired domain. Otherwise how can you tell the difference between an image of "red green blue" and one of "green blue red"? It's too much work to imagine the underlying words in order to decide whether two potentially radically different avatars are the same. It's easy enough to try both though 🤷🏻‍♂️