VLM

Also known as: VLM, Vision Language Model

A vision-language model (VLM) takes an image (and usually a text prompt) and generates text. Typical stack: a vision encoder turns the image into a sequence of vectors, a projector maps those vectors into the language model’s embedding space, then an instruct LLM continues the sequence with next-token prediction.

What it outputs

Captions, answers to visual questions, OCR-style reads, or structured text such as JSON boxes. The boxes are still text tokens (digits and punctuation), not a separate detection head like YOLO.

Multimodal is the broader class (image, audio, video, text). A VLM is the common image+text instance used at inference for chat-over-pixels.

← Back to AI Glossary