Video Depth

The Video Depth node turns a video into a depth video: near is bright, far is dark. It estimates the depth of every frame and renders the depth as an MP4 of the same clip. Feed the grayscale result into depth-guided generation, relighting or compositing, or choose a colormap for a stylised look.

When to use it

• You need a depth pass of a clip for compositing, for example to place a layer between the foreground and the background.
• You want a depth video as a guide for another generation or for relighting.
• You want a stylised depth look, such as a Turbo or Inferno colormap, for a title or a social clip.
• You want to compare the original clip and its depth side by side.

Quick start

Inputs

Outputs

Settings

The settings panel shows only the settings the chosen model takes. A setting the model does not take is ignored, and a colormap the model cannot render becomes Grayscale. A run is never refused for either, so a model that a field mapping switches still runs.

Models

Credits

The price depends on the model and on the length of the input clip. The length is rounded up to the next tier: 15 seconds, 30 seconds, 1 minute, 2 minutes or 5 minutes.

• How the tier is chosen. A 5-second clip and a 15-second clip are both in the 15s tier. A clip of 15.01 seconds is in the 30s tier, and a 61-second clip is in the 2min tier.
• The clip is measured before the run. A clip that cannot be read is refused before any credits are reserved. A run is never priced at a guessed length.
• The price on the node. Until the length of a connected video is known, the node shows the 15s tier. The price updates once a video with a known length is connected.
• Depth Anything (per frame). This model accepts clips up to 80 seconds, so its 2min tier is the most a run can cost.

Tips

• Keep Grayscale for other nodes. Grayscale works best as the input of another node. Colormaps are for viewing.
• Preview with Small, finish with Large. Use Small (fastest) to check a clip quickly, and Large (best) for the final pass.
• Trim first. The price follows the length of the input clip. Cut the clip to the part you need with Trim Video before Video Depth.
• Choose Video Depth Anything for moving shots. Depth Anything (per frame) works on each frame alone, so the depth can flicker between frames.

Troubleshooting

From the API

POST /v1/video-depth takes videoUrl, with optional provider, modelSize, colormap, resolution and sideBySide. provider is video-depth-anything (the default) or depth-anything-frames, and modelSize is small, base or large (the default). colormap is grayscale, turbo, inferno, magma or viridis, and resolution is auto, 360p, 480p, 720p or 1080p.

POST /v1/video-depth
Content-Type: application/json

{
  "videoUrl": "https://example.com/clip.mp4",
  "provider": "video-depth-anything",
  "modelSize": "large",
  "colormap": "grayscale"
}

The answer is { jobId }, and the finished job's output is { videoUrl, thumbnailUrl }. A setting the model does not take is ignored, and a colormap it cannot render becomes grayscale. A value that is not in the lists above is a 400 validation_error. The clip is measured before anything is created: 400 video_too_long and 400 video_unreadable reserve nothing.

With the SDK, run the node by its type:

const result = await client.nodes.run("video-depth", {
  videoUrl: "https://example.com/clip.mp4",
  colormap: "grayscale",
})

With the CLI:

nodaro nodes run video-depth --param videoUrl=https://example.com/clip.mp4 --watch

From an AI assistant, the MCP tool is video_depth, with video_url or video_asset_id and the optional provider, model_size, colormap, resolution and side_by_side. See Run a single node, the SDK nodes reference, the CLI commands and the MCP tools.