Analyze an image with custom vision model.

[razvanab]

Wherever I choose an image to analyze within the Jan app, Jan always asks me if I want to install the Jan V2 VL Model instead of using the vision model I choose to do so.

[wellkilo]

Jan decides whether the image picker can be used from the selected model's declared vision capability. The Jan-v2-VL prompt is the fallback shown when the currently selected model is not recognized as vision-capable; it does not mean Jan must use Jan-v2-VL.

For a custom/remote model, open its provider settings, edit that model, and enable Vision under Model Capabilities. Then select that same model again before attaching the image. Jan's built-in providers can populate capabilities automatically, but a custom OpenAI-compatible endpoint generally needs this metadata set manually.

Two details are worth checking:

• The backend really must accept OpenAI-style multimodal message content (an image_url content part). Marking the capability only exposes the UI; it cannot add vision support to a text-only endpoint.
• For a local GGUF vision model, import the matching multimodal projector/mmproj as part of the model import. A language GGUF without its projector cannot process the image even if the flag is enabled.

The current file-upload documentation states that Add Images appears only when the active model has vision, and the provider UI explicitly supports manually editing model capabilities: file upload docs.

So the short answer is: you do not need to install Jan-v2-VL; declare Vision on the custom model and make sure its endpoint/model files genuinely support multimodal input.

[razvanab]

I see
Thanks