Video Depth Map node: bake a depth clip or pass depth into Video Render

Estimate depth frame by frame in the open browser, or carry the depth operation into a downstream Video Render composition.

Where: Flowgen -> Add -> Video & Motion -> AI Edit -> Video Depth Map

Key ideas

A runnable local node: Video Depth Map now has a Run action. It decodes the source clip, estimates depth, and bakes a source-FPS output in the browser. It is no longer only a non-running instruction card.
Two delivery paths: Run this node for a standalone depth clip, or keep its depth-pipeline settings connected to Video Render for composed output. The standalone local route and downstream compositor are separate execution paths.
Three map views: Grayscale is white-near/black-far, Inverted reverses it, and Turbo is a diagnostic false-color view. Choose a meaningful orientation for downstream depth consumers rather than a visually attractive diagnostic map.

Connect a source clip to video_in.
Open the node’s settings on the card or sidebar and choose colormap: grayscale, inverted, or turbo.
Run the node in the open browser when you need a baked clip. Watch progress and wait for completion before using the result.
Inspect the rendered depth video and frame count/FPS information. A settings change invalidates the old result preview until the current settings are run.
For composed delivery, wire video_out into Video Render or the supported depth chain and configure final output on the renderer.
Review frames around cuts and fine edges; a successful encode does not prove temporally stable depth.

One video_in source and video_out output; no text prompt or paid image-generation model is required.
Complete opParams: { colormap: "grayscale" | "inverted" | "turbo" }.
Local processing detects source frame rate and saves an output artifact when project storage is available.
The card can show the finished result; while no valid current result exists it shows the source.

Use Grayscale for compositing and most downstream depth work.
Use Turbo to inspect separation, not as numeric depth for blur.
Test a short clip before a long render, especially around hair, glass, motion blur, and scene cuts.

Local baking requires an open browser and readable source media; it cannot run in a server-only/headless flow.
Parameters are static across the clip; there are no depth keyframes.
Estimated depth can shimmer around motion blur, transparent surfaces, fine hair, or scene cuts.

Run the local node in the browser, or use the appropriate downstream rendering workflow. Local model execution is not a headless provider API.