Vision Model
Model Visi
A model that reads images, turning them into something a language model can reason over.
A vision model takes an image and produces a representation a transformer can work with. In the common design, an image encoder produces a grid of vectors that are inserted into the token stream, so the rest of the model treats image content much like text.
This is why a modern multimodal model can read a document, reason about a chart, or describe a screenshot without a separate pipeline for each task. The capability comes from the same scaling that produced the language ability.
The practical limits are worth knowing: small text in an image is hard, spatial reasoning is weaker than object naming, and reading a dense table from a photo is a different task from reading a table in a file. When accuracy matters, giving the model the file is usually better than giving it a photo of the file.
Related terms
Multimodal
A model that handles more than one kind of input or output: text, image, audio, video.
Diffusion Model
The architecture behind most image and video generators.
Context Window
Everything the model sees on one request. Finite, and it decides what fits.
OCR
Turning an image of text into text you can select and search.
Embedding
A list of numbers that places text near other text with similar meaning.
Want this working inside your team?
It starts with one conversation about your workflow.