Vision-language models

Vision language models (VLMs) process images and text within a single architecture. The standard pattern: an image encoder (CLIP ViT, SigLIP) produces visual embeddings, a projection layer aligns them to the language model embedding space, and the language model decodes them into text.