search

dcramer / gamegame

https://gamegame.ai
Apache License 2.0
8 stars 3 forks source link

Strategy to embed images #6

Open dcramer opened 2 months ago

dcramer commented 2 months ago

There's a lot of visuals in rulebooks that'd be great to reference. Ignoring whether we can textual represent them, the marker parser we're using will actually transform text and extract images at the same time. I believe we could design a system that lets us easily reference those in outputs (and embed them).

tian-yi commented 2 months ago

Yes, that's exactly copali for. Copali is used to embed the images. and direct semantically search the images using text. Also, the cool part is that you can then feed the retrieved image and text into a multi-modal LLMs like gpt-4o, gemini etc. they will answer you questions using both.