Scene Detection

Sieve's Scene Detection is an automated scene detection API that identifies scene transitions in video content.

For pricing, click here.

For further information, click here.

Parameters

• video: Input video file to process (minimum 3 seconds; max 30 mins for backend=turbo)
• backend: (default: "base") Choose between "base" (high quality) or "turbo" (faster processing). For details, see Backends.*
• start_time (default: 0): Start time in seconds to begin processing
• end_time (default: -1.0): End time in seconds to stop processing (-1 for end of video)
• return_scenes (default: false): Whether to return individual video clips for each detected scene
• min_scene_duration (default: 0.1): Minimum length in seconds for a clip to be considered a scene and included in the output (must be >0)
• threshold (default: 1.0): Adjusts detection sensitivity. Higher values result in fewer scene transitions. Recommended to increase/decrease in increments of 0.1, which is equivalent to ~10% increase/decrease in the sensitivity of the scene detection
• transition_merge_gap (default: 0.1): Maximum gap in seconds between frames to be considered part of the same scene transition (i.e., if transition_merge_gap=0.5 and there are detected transitions <0.5s apart, they will be grouped together)

• For backend options, see Backends.

Important Notes

• Scene detection sensitivity is automatically adjusted based on video characteristics; threshold further manipulates this sensitivity
• The algorithms struggle to identify slow fade transitions between structurally similar scenes; backend="base" will identify fades more accurately
• Non-H.264 videos are automatically re-encoded to H.264 using hardware acceleration, preserving the original resolution and aspect ratio, though slight quality loss from compression may occur

Output

If return_scenes is False:

• A single dictionary of scenes containing the following for each detected scene:
  • scene_index: Index of the scene
  • start_time: Start time of the scene in seconds
  • end_time: End time of the scene in seconds
  • start_frame: The starting frame number of the scene
  • end_frame: The ending frame number of the scene
  • duration: The length (in seconds) of the scene

If return_scenes is True:

• A tuple (clip, metadata) where:
  • clip: A sieve.File object representing the rendered scene
  • metadata: A dictionary with the keys listed above for each scene

Backends

Backend	Description	Speed	Quality	Price (per minute)* Up to 4K return_scenes=false	Price (per minute)* Up to 4K return_scenes=true	Price (per minute)* 4K+ return_scenes=false	Price (per minute)* 4K+ return_scenes=true
base	Sieve's premier quality scene detection solution. Accurately identifies scene transitions including fades and complex transitions.	⭐️⭐️	⭐️⭐️⭐️⭐️	$0.05	$0.10	$0.10	$0.20
turbo	A fast solution that works well for most videos with clear scene transitions. Approximately 3x faster than base.	⭐️⭐️⭐️⭐️	⭐️⭐️⭐️	$0.02	$0.07	$0.04	$0.14

*Pricing assumes 30 fps video content.

backend="base"	backend="turbo"

For longer (>30 min) videos, or videos with subtle transitions or fade effects, we recommend using base. If processing speed is more important than detecting every transition, we recommend using turbo.

Notes

• Discounts are available for high volume users. Please reach out to sales@sievedata.com or via Discord for more information.