Generate Video Depth
Generate a temporally-consistent depth video from footage (Video Depth Anything). Spends AI tokens, scaled by input duration.
Category: AI Menu path: AI > Generate Video Depth
Ports
| Port | Type | Direction | Description |
|---|---|---|---|
video_in | imageRgba16f | input | Required — connect a Video Source (or another AI video node). The actual video file is uploaded for processing; effects chains aren't supported yet — bake them with an export first. |
out | imageRgba16f | output | The depth video, frame-aligned with the input. |
Parameters
AI controls live in the Generate panel; standard rows expose Video Source-style playback params.
| Param | Type | Default | Description |
|---|---|---|---|
model | string | (first available) | Single curated model (Video Depth Anything). |
resultPath | string | "" | Hidden — the saved depth clip under generated/. |
How It Works
The input video uploads from your machine, the job runs remotely (async — status streams into the button), and the depth video downloads into generated/. Unlike running Generate Depth per frame, the video model is temporally consistent — depth doesn't flicker between frames. Cost scales with the input's duration; the preview shows it once the input is probed. Failures refund automatically.
Usage Examples
Depth-driven atmosphere
Footage → Video Depth → use the depth video as a Mix mask over a graded copy of the footage — fog, color, or blur that falls off with distance, tracking the camera through the whole clip.
Frame-accurate displacement
Wire the depth output into UVRemap's uv_map against the original footage for parallax/displacement looks.
Tips
- Match the layer timing of the depth clip to the source footage (same start frame and speed) so they stay frame-aligned.
- Trim the source first — you pay per second of input.
Related Nodes
- GenerateDepth — single-frame version
- VideoSource — playback semantics
- UVRemap, Mix — typical depth consumers