Is it possible to get an AI-based, repeatable pubkey -> memorable profile image generator so that every time the app asks to generate an image based on a key, it gets the exact same one?

The goal is to improve our visualization of the user's key so that you can immediately know if you are talking with the same person or an impostor.

It must be an AI because humans cannot make enough image layer combinations to represent every key, like on Robohash.

Each user should also create a nonce so that I might be a rabbit for one user or an astronaut for another. In that way, we minimize impersonation attacks.

1. It must generate memorable images.

2. It must have a large set of elements to combine so that every possible key can be represented without being too similar to any other key.

3. It must be repeatable since each client needs to run it locally over the pubkey.

🤔

Reply to this note

Please Login to reply.

Discussion

The rate that we fund new uses for AI.... 🤯

Very cool idea 👍

Google has an emoji combiner idk how many combinations you need, but it's something that is small and yet recognizable

Kinda like with the pubkey color thing in nostrudel?

yep, or the robohash in other clients.

The only issue is that those solutions are not memorizable.

Wow that's a cool idea!

I'd wondered about this a while back. Isn't that sort of what samurai did with the robot paynyms? Not as memorable as one would want, but I couldn't find documentation on how it's done. A good twist imo would be to generate meme remixes. Memorable and language agnostic 12 word scheme

*U* A+ smart k'O;.;O'ky/*****LFg'O'*****Y'O'DL/*\************************************************************************************************************************/*\ya/>]

You don't need to use one style. Dicebear.com have 30 avatar styles. Combining all of them can be more memorable for human.