PDF Page to Image Conversion for Multimodal LLM Input - Necessary?

[buddypia]

Hi team,

I've been working on a feature that converts entire PDF pages or specific portions of pages into images. This would enable the use of these images as input for multi-modal LLMs.

My question is: Do we actually need the multi-modal LLM to provide descriptions/explanations based on these images? I've almost finished building the image conversion part, so if the descriptive capability from the LLM is desired, I can put together a PR. Let me know your thoughts.

Thanks!