sidhant
@sidhant
I'm so obsessed with the idea that generative AI is the future of video games. the only problem is unit economics. gamers expect to spend $20-$60 one time for hours and hours of entertainment. nobody wants to pay a monthly sub. so: edge llms. run the game logic on device. no api fees. I've been playing around with this idea, my current experiment uses gemma3n running on my iphone to generate svg graphics. here’s it drawing a “puppy.” sadly this attempt ended up being a dead end. does anyone have better ideas for representing visual output from an edge llm? maybe some kind of json shape primitive -> sprite renderer thing?
4 replies
2 recasts
40 reactions
sidhant
@sidhant
even though the drawings are shit, I did discover that this thing is really good at following instructions. it reliably gives me JSON for every response, which allowed me to do this dual chat + SVG interface!
0 reply
0 recast
1 reaction
sidhant
@sidhant
update: I'm getting way better results with the JSON shape renderer approach. might actually be able to get this to work with better prompting!
0 reply
1 recast
0 reaction
Pretty Blocks
@prettyblocks
I don't think LLMs speak SVG in this way. Might be better to get a lightweight image gen model (diffusion or other). Then just trace the output with a jpg/png -> svg tracing tool. You can also generate sprite sheets on the fly.
1 reply
0 recast
0 reaction
Denver Dsouza
@denverjude
Why won’t people pay for a sub though? I pay for PS Plus like many others. Edge LLMs might not be efficient for graphics heavy rendering. Cool idea to explore nonetheless.
1 reply
0 recast
0 reaction
