Caption or read text from an image
run with ollama
// what it does
Gemma 4 is natively multimodal: put an image path in the prompt and it describes, captions, or OCRs the picture. All sizes accept images; E2B, E4B, and 12B additionally handle audio and video frames. Point the path at a real file on disk.
// shell
$ ollama run gemma4 "describe ./chart.png"// gotcha
Place the image reference before the instruction. Gemma 4 is trained with media tokens ahead of text, so text-first prompts measurably weaken visual grounding even though they still run.