video demo: https://www.youtube.com/watch?v=EK7z-HltssE
api usage: Unfortunately Gemini Flash 2.0 image Gen has really tight API limits that don't scale up for paid tiers -- if anyone on the team sees this and wants to help me out, my email is in my profile or you can reach me at hello@lildigi.me
level stats: For the speedrunners out there, the fastest glitchless times I've seen around 17-18s.
stack: cloudflare pages + aws lambda + gemini
happy to answer any questions!
That being said, the resolution is such that saying messy hair, full beard, and black glasses, would pretty much get it.
I think it’s simply a matter of it assumed a person, found glasses, and a bit of hair above and below and filled in logically.
On prompting, you can get most of the way there in AI studio on Gemini 2.0 Flash (Image Generation) Experimental by uploading a picture and asking for "a high quality detailed pixel art sprite of this character." Most of the backend annoyance here was iterating to improve prompt adherence (characters not facing the same way, outfits changing between frames, etc).
This is pretty cool
Apologies for the inconvenience! If anyone on the gemini team can help out, please lmk (email in bio).
Best in the system now is 17.1s and it must have been a near perfect run.
You may be able to take advantage of CORS so the Gemini call is client-side and your backend never sees the key. Simon Willison did this for Claude and wrote it up here: https://simonwillison.net/2024/Aug/23/anthropic-dangerous-di...
Very cool idea, thanks for sharing!
iguana•9mo ago
sunnyba•9mo ago