Knowledge MaxxingVision-language model
A vision-language model is trained to understand both images and text together, so it can answer questions about an image or describe it in natural language.
Prerequisite: Multimodal model
What matters
It works by aligning image and text representations in a shared space, so a description and the image it describes end up close together, letting the model reason across both.
Example
A vision-language model can look at a photo of a receipt and answer 'How much was the total?' by reading the text within the image and reasoning about it.
Common misconception
A vision-language model is not simply an image captioner bolted onto a separate text model; it is trained so both modalities share one underlying representation.
Recall question
What lets a vision-language model reason jointly about an image and a related question?