Show HN: Generate Textual Descriptions for Images

cool, though it hallucinate a bunch of details

also it's way too expensive for my use case sadly

do you use a modified CLIP model or something else entirely?