Extract depth maps with Depth Anything V2. Turn any image or video into a per-frame structure signal for AI video motion transfer. Try free. Extract a per-frame depth map from a video or image using Depth Anything V2, for use as a motion/structure control signal in AI video generation.
Extract a per-frame depth map from a video or image using Depth Anything V2, for use as a motion/structure control signal in AI video generation.
About this model
Depth Anything V2 extracts a precise per-pixel depth map from a single image or every frame of a video, producing a grayscale map where brightness encodes relative distance from the camera. For video input, frames are processed in batches on GPU and reassembled at the source resolution and frame rate, giving a temporally aligned depth sequence. This is commonly used as a motion and structure control signal for AI video generation models such as Seedance 2 and Seedance 2.5, where a driving video's depth map guides camera movement and scene geometry while the appearance is generated fresh. It is equally useful for 3D-aware compositing, relighting, background removal, and any pipeline that needs a stable geometric reference from ordinary RGB footage.
Cost analysis
| Provider | Cost | Notes |
|---|---|---|
| muapiapp | $0.04/sec of video, or $0.05 flat per image | Pay per second of input video processed, or a flat rate for single-image depth maps, with no subscription. |
| Fal.ai | Not available | This model variant is not listed on Fal.ai. |
| Replicate | Not available | This model variant is not listed on Replicate. |
Pay per second of input video processed, or a flat rate for single-image depth maps, with no subscription.
This model variant is not listed on Fal.ai.
This model variant is not listed on Replicate.
** Competitor pricing is estimated based on similar model architectures and usage tiers.
Configuration schema
| Parameter | Type | Description | Default |
|---|---|---|---|
| Video URL | string | The URL of the driving video to extract a per-frame depth map from. Provide exactly one of video_url or image_url. | https://storage.googleapis.com/gtv-videos-bucket/sample/ForBiggerBlazes.mp4 |
| Image URL | string | The URL of a single image to generate a depth map from. Provide exactly one of video_url or image_url. | https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/seedream-5.0-edit.jpg |
The URL of the driving video to extract a per-frame depth map from. Provide exactly one of video_url or image_url.
https://storage.googleapis.com/gtv-videos-bucket/sample/ForBiggerBlazes.mp4The URL of a single image to generate a depth map from. Provide exactly one of video_url or image_url.
https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/seedream-5.0-edit.jpgDeveloper documentation
video_url (a driving video for per-frame depth extraction) or image_url (a single frame for a one-off depth map)./api/v1/depth-anything with a publicly accessible URL.webhook_url to be notified on completion.curl -X POST "https://api.muapi.ai/api/v1/depth-anything" \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"video_url": "https://example.com/driving-video.mp4"
}'
Frequently asked
It is a monocular depth estimation model that produces a grayscale depth map from a single image or every frame of a video, where pixel brightness represents relative distance from the camera.
Provide exactly one. Use video_url to extract a per-frame depth video (e.g. for motion-transfer control signals), or image_url for a single-frame depth map.
A video input returns a grayscale depth-map video at the same resolution and frame rate as the source. An image input returns a single grayscale depth-map PNG.
Seedance's motion-transfer workflows accept a depth video as a structure/motion control signal. Extract the depth map from a driving video with this API, then pass it into Seedance to guide camera movement and scene geometry while generating new appearance.
The API accepts standard web-hosted image and video files. Very long videos increase processing time since every frame is run through the depth model; keep driving videos to the length you actually need for the downstream generation.
Video input is billed at $0.04 per second of input video. A single image input is billed at a flat $0.05 per request.