This was a fun experiment to try PaliGemma (open vision-language model). I found that PaliGemma performed better than Gemini Flash for this type of specific image task, especially around latency. (~0.9 seconds for PaliGemma inference on a VM, vs. 3-4 seconds for Gemini Flash.) Would love feedback on ways to potentially improve this setup.
1 comment
Show HN: Papermusic (draw an instrument, then play it) | Heykuki News