Images
Send a screenshot or image with your question. With an image-capable model, the model receives the image itself. Images returned by tools, such as a browser screenshot or a PNG opened with read, reach the model too; the vision model described below handles them when your main model cannot see images.
In the editor, type an image path with @:
What is wrong with this layout? @screenshots/search.pngYou can also paste an image with Ctrl+V, where clipboard images are supported, or drag an image file onto the terminal. Pasted and dragged images appear as numbered image markers in your draft.
For print mode, use -image; repeat it to send more files:
tatu -p "Explain this error dialog." -image screenshot.pngSize and format
Section titled “Size and format”Image files can be PNG, JPEG, GIF, WebP or BMP. Tatu shrinks images to fit within 2000 × 2000 pixels and 4,500,000 bytes of base64-encoded data (about 4.5 MB after encoding). It can recompress or convert them to fit. Files larger than 50 MiB are rejected before loading.
When your model cannot see images
Section titled “When your model cannot see images”Use /settings vision to choose an image-capable model that describes images for your main model. You need credentials for its provider. The description becomes the text your main model sees, and its cost appears in /cost.
Without a vision model, an image becomes only a short note for a model that cannot accept it. Attaching an image does not make that model able to see it.