3 ms·
> there is no way of rasterizing a vector image to 16x16 and preserving anything but a blurry mess. Deep learning suggests differently?
by nox101 2y ago
> there is no way of rasterizing a vector image to 16x16 and preserving anything but a blurry mess.
Deep learning suggests differently?
- 8n4vidtmkvmk 2y agoSo instead of 5kB bitmap we'll do a 768B vector + a 5GB diffusion model, plus waste 20 seconds just to render it.
- nox101 2y agoYou got me on the model size but speed wise there are algos that run fast https://fastsdxl.ai/ https://fastsdxl.ai/
- erhaetherth 2y agoThose are pretty fast but I suspect spending nearly any amount of time running some model to spit out an icon is too much time. I should be able to load a directory with 1000s or 10s of thousands of files extremely quickly.