visual-prompting

ART fine-tunes multimodal LLMs through images

ART tunes frozen multimodal LLMs by optimizing a single image instead of model weights.