DOCUMENTATION / Using Noullama
Images & vision
Load a multimodal projector and ask typed questions about images.
Prepare the model and projector
Image input requires both an image-capable model and its compatible multimodal projector.
- Place the projector GGUF, whose name contains
mmproj, in the same folder as the model. The Hub lists projectors beside the model weights. - Turn on Vision in the model bar, or the projector switch under Runtime → Model.
- Load the model with the projector enabled.
Selecting a model in another folder turns the projector off. Check the Vision setting when switching models.
Add images to a decision
In Playground, use Add images, drag files onto the State box, or paste them. Choose your questions and run the decision normally.
You can leave the state text empty to ask about the images alone.
Formats and app limits
PNG, JPEG, GIF and BMP are sent unchanged. Other formats the app can display, such as WebP, are converted to PNG first.
- Each image can be up to 20 MB.
- All images together can be up to 60 MB.
Send images through the API
Native clients can send an images array of data:image/...;base64,... URLs to /v1/systemone when the model is loaded with its projector.