Vision & images
Drop an image into a room and ask about it — uncensored vision with no refusals.
Uncensored vision#
Select darkroom/vision and you can attach an image to any message. The model reads it and answers questions about it the same way it answers text — directly, with no refusals and no "I can't analyse this image."
How to use it#
- 01Switch the model to
darkroom/vision. - 02Click the image button in the composer (or drag an image in) and add your question.
- 03Send. The model sees the image alongside your prompt.
What it's good for#
- —Reading screenshots, documents, and diagrams.
- —Describing or analysing photos other models refuse to look at.
- —Pulling text and structure out of an image (OCR-style tasks).
Notes#
- —Images are processed inside the sealed room like any other input — decrypted only in the enclave, never logged.
- —You can mix text-only and image turns in the same room; only the vision model will actually see attached images.