Depth Anything V3 Video API Documentation

Depth Anything V3 Video API Documentation

Playground

Depth Anything V3 Video turns any video into a temporally consistent depth map video with no flicker, ideal for replicating camera moves and motion with depth-controlled video generation. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

Features

Depth Anything V3 Video converts an input video into a temporally consistent depth-map video using Depth Anything V3. Depth is normalized across the full clip rather than independently frame by frame, helping maintain stable depth values as subjects move, the camera changes position, and the scene evolves.

The generated depth video preserves the input resolution, duration and audio track, with output frame rate preserved up to 30 fps, making it suitable for depth-conditioned video generation, compositing, relighting, parallax, and other motion-aware workflows.

Why Choose This?

• Temporally consistent depth
  Depth is normalized across the entire video to reduce frame-to-frame brightness shifts and depth flicker.

• Stable through fast motion
  Depth is estimated with temporal context across neighboring frames, so structure holds up through fast camera moves and motion blur instead of changing from frame to frame.

• Stable subject depth
  Subjects can maintain more consistent depth values as the camera moves or the scene changes.

• Clear subject separation
  Preserve silhouettes, thin structures, people, hands, and background geometry for downstream control workflows.

• Control-workflow ready
  Use the default grayscale output directly in depth-conditioned video generation and ControlNet-style pipelines.

• Multiple colormaps
  Choose grayscale for depth workflows or inferno and turbo for visualized depth output.

• Long-video support
  Process up to 600 seconds of video in a single request.

• Original resolution and frame rate
  Output follows the input video resolution and preserves frame rate up to 30 fps.

Parameters

How to Use

1. Provide an input video — Upload a video or provide its URL.
2. Choose a colormap optional — Use grayscale for generation and compositing workflows, or inferno / turbo for visual visualization.
3. Submit — Generate the depth-map video.
4. Use the result — Feed the output into depth-controlled generation, relighting, compositing, parallax, or other depth-aware workflows.

Pricing

Pricing is based on input video duration.

The rate is $0.005 per billed second.

Billed duration is rounded up to the next whole second, with a minimum billed duration of 3 seconds and a maximum billed duration of 600 seconds.

colormap does not add a separate charge.

Best Use Cases

• Depth-conditioned video generation — Use the depth video as structural guidance for recreating motion and camera behavior with new visual content.
• Motion and camera replication — Extract depth from a reference clip and use it to guide another video-generation workflow.
• VFX and compositing — Apply depth-based masking, grading, fog, volumetric effects, and layering.
• Relighting — Use a temporally stable depth pass for depth-aware lighting adjustments.
• Depth of field — Create foreground and background separation for focus and blur effects.
• 2.5D and parallax effects — Use depth information to create spatial movement from existing footage.
• 3D-aware editing — Apply depth-driven transformations while preserving scene structure over time.
• AR occlusion — Use scene depth to help virtual objects pass in front of or behind real-world subjects.

Pro Tips

• Use grayscale when the depth video will be passed into another model or compositing workflow.
• Use inferno or turbo when the output is primarily for visualization or inspection.
• In grayscale mode, white represents surfaces closer to the camera and black represents surfaces farther away.
• Treat the result as relative depth, not real-world metric distance.
• Depth consistency across the full clip makes the output more suitable for motion-aware control workflows than independently processed frame-by-frame depth maps.
• Videos above 30 fps are returned at 30 fps while preserving the same duration.
• The output keeps the source audio, so the depth video stays in sync when you lay it back over the original edit.

Notes

• video is required.
• colormap defaults to grayscale.
• Supported colormaps are grayscale, inferno, and turbo.
• Up to 600 seconds of input video are processed.
• Longer videos are processed only up to the first 600 seconds.
• Output resolution matches the input video resolution.
• Output frame rate matches the source up to 30 fps.
• Inputs above 30 fps are returned at 30 fps with the same duration.
• The source audio track is kept in the output video.
• Depth is relative rather than metric.
• In grayscale mode, near surfaces are white and far surfaces are black.

Related Models

• Depth Anything Image — Generate grayscale depth maps from individual images.
• Depth Anything Video — Generate depth maps from video input.
• Depth Anything V3 Image — Generate detailed image depth maps with Depth Anything V3.
• Depth Anything V3 Video — Generate temporally consistent depth-map videos with Depth Anything V3.

Authentication

For authentication details, please refer to the Authentication Guide.

API Endpoints

Submit Task & Query Result

set -euo pipefail

export WAVESPEED_API_KEY="your-api-key"

REQUEST_BODY=$(cat <<'JSON'
{
  "video": "https://interactive-examples.mdn.mozilla.net/media/cc0-videos/flower.mp4",
  "colormap": "grayscale"
}
JSON
)

# 1. Submit the prediction.
SUBMIT_RESPONSE=$(curl --silent --show-error --fail-with-body \
  -X POST "https://api.wavespeed.ai/api/v3/wavespeed-ai/depth-anything-v3/video" \
  -H "Authorization: Bearer ${WAVESPEED_API_KEY}" \
  -H "Content-Type: application/json" \
  -d "${REQUEST_BODY}")

TASK=$(printf '%s' "${SUBMIT_RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
PREDICTION_ID=$(printf '%s' "${TASK}" | jq -r '.id // empty')
if [ -z "${PREDICTION_ID}" ]; then
  printf 'Submission response did not contain a prediction id
' >&2
  exit 1
fi
RESULT_URL="https://api.wavespeed.ai/api/v3/predictions/${PREDICTION_ID}/result"

# 2. Poll until the prediction finishes.
while true; do
  RESPONSE=$(curl --silent --show-error --fail-with-body \
    "${RESULT_URL}" \
    -H "Authorization: Bearer ${WAVESPEED_API_KEY}")
  RESULT=$(printf '%s' "${RESPONSE}" | jq 'if type == "object" and has("data") then .data else . end')
  STATUS=$(printf '%s' "${RESULT}" | jq -r '.status // empty')

  case "${STATUS}" in
    completed) printf '%s\n' "${RESULT}" | jq '.outputs'; break ;;
    failed|cancelled|timeout|deleted) printf '%s\n' "${RESULT}" | jq . >&2; exit 1 ;;
    *) sleep 2 ;;
  esac
done

Parameters

Task Submission Parameters

Request Parameters

Response Parameters

Result Request Parameters

Result Response Parameters