Input Image HTTP

1 VersionInput

Exposes a public image ingress that accepts client-pushed bytes over HTTP POST or WebSocket, decodes them, and emits an Image paired with a per-upload session-id String. Use it when a deployment needs to take images from outside the graph.

How it fits

Typical backends

Stateless detection endpoint — push an image, get detections as JSON.

OCR as a service — push an image, return recognised text as JSON.

Classify an uploaded image and stream the annotated result back over HTTP.

Caveats

Hard constraintClients must send raw image bytes; form-data uploads are not parsed. transport fixes the open channels and max_image_bytes rejects oversized uploads before decoding, and neither can change at runtime.
I/O contractEach upload gets a fresh session-id emitted alongside the Image so downstream can correlate the inference result back to the request; WebSocket reconnects are silent and only the per-frame id ties messages together.
Fallbackbit_depth_mode decides how higher-bit-depth source pixels are mapped down to 8-bit, and an unrecognised mode quietly falls back to linear scaling while already-8-bit input is untouched.
LatencyTurning drop_frames off makes the HTTP response wait until the pushed frame has landed downstream, so a slow consumer slows every POST.
State lifetimedrop_frames is the only hot-swappable knob; transport, max_image_bytes, and bit_depth_mode bind once at startup.

Versions

7adf1b57latestdefault
Rebuild against the current pipeml
7/29/2026