openeye run

    Run inference on an image, a directory of images, or stdin, and print JSON results. The model must be pulled first.

    openeye run yolov8 photo.jpg
openeye run yolov8 photo.jpg --pretty
openeye run yolov8 photo.jpg --confidence 0.5
openeye run depth-anything scene.png
openeye run grounding-dino image.jpg --prompt "red cup"
openeye run yolov8 photo.jpg --visualize
openeye run yolov8 photo.jpg --backend onnx
openeye run yolov8 ./images/ -o results.json

    Arguments

    ArgumentDescription
    MODELModel name (e.g. yolov8)
    IMAGEImage path, directory, or - for stdin. Omit it when piping into the command

    Options

    FlagDescription
    --prompt, -p <text>Text prompt for open-vocabulary models (e.g. grounding-dino)
    --confidence, -c <float>Minimum confidence threshold (0.0–1.0)
    --output, -o <path>Write JSON results to a file (indented)
    --prettyPrint a human-readable, colored summary to stderr; JSON still goes to stdout
    --visualizeSave an annotated image with bounding boxes or depth map
    --backend, -b <name>Runtime backend: onnx or tensorrt (selects that variant)
    --variant <name>Use a specific model variant

    A directory runs every image in it and prints a JSON array. Images that fail are skipped with a warning.

    Unix Pipeline

    Output is a single line of JSON on stdout, so it composes with standard tools:

    openeye run yolov8 photo.jpg | jq '.objects[] | {label, confidence}'
openeye run yolov8 photo.jpg | jq '.objects | length'
cat photo.jpg | openeye run yolov8 -

    Piping one run into another merges their results, for example detections plus a depth map:

    openeye run yolov8 photo.jpg | openeye run depth-anything -
    ← openeye pullopeneye serve →