Analyze Image (Ollama VLM)

@ollama1 VersionClassify

Sends one or more images plus a String prompt to a vision-language model served by a local Ollama sidecar and returns the model's reply as a String. The image input is a oneof spanning a single image, an image-with-label, a batch, or labeled batch.

How it fits

Typical backends

Caption or visually reason over frames pulled from a video file and emit the model's answer as JSON for a downstream service.

Describe an image fetched over HTTP and read the answer aloud through a text-to-speech component.

Caveats

CompatibilityA local Ollama sidecar must be reachable and the selected model is pulled at startup if not already cached; the first deployment of an uncached model blocks on that download before any request is served.
State lifetimetemperature, num_predict and num_ctx are hot-reloadable per request, while the model, image encoding and concurrency wiring are read once at startup and fixed for the life of the deployment.
Hard constraintmax_parallel sets the request pool size at startup, so raising throughput requires a redeploy rather than a config update; an out-of-range hot update to a tunable is rejected and the previous value is kept.
Parameter interactionnum_ctx left at its zero sentinel keeps the model's own default context window; raise it above the prompt plus image tokens plus the generation budget when long outputs are truncated.
CostEach image is re-encoded to image_format.jpeg or image_format.png and base64-embedded into the request body, so large batches inflate payload size and per-call latency.

Versions

c60cd690latestdefault
Automated release
8/2/2026