Skip to content

Video Frame Analyzer

FieldValue
Namevideo-frame-analyzer
Categorydata-pipeline
Complexitymoderate
Tagssidecar, ffmpeg, video-processing, llm-synthesis, shared-volume
Authorrandybias
Min Version0.7.0

Extract frames from a video using an ffmpeg sidecar and analyze them with Claude Vision. This is the simpler sidecar scaffold for video processing, demonstrating the shared volume data flow pattern. The engine stages the input video, the sidecar extracts frames, and the engine reads the results — all via an emptyDir volume mounted at /shared.

This scaffold includes a sidecar container:

FieldValue
Nameffmpeg
Imageghcr.io/randybias/tentacular-ffmpeg-sidecar:v1.0.0
Port9000
Health checkGET /health
CommunicationShared emptyDir volume at /shared/

The ffmpeg sidecar uses a custom image that wraps ffmpeg with an HTTP API. The engine writes the input video to /shared/input/video.mp4, calls the sidecar’s /extract-frames endpoint, and reads extracted JPEG frames from /shared/output/.

Contrast with doc-converter which uses pure HTTP body data flow for smaller text payloads.

ingest-video → extract-frames → analyze-frames
NodePurpose
ingest-videoFetch video from URL and stage to /shared/input/video.mp4
extract-framesCall ffmpeg sidecar to extract frames at configured FPS
analyze-framesSend frames to Claude Vision as base64-encoded images for analysis
  • manual only
ServiceTypeRequired
Anthropic APIExternalYes (Vision analysis)
Video sourceExternalYes
KeyDefaultDescription
timeout300sPer-node timeout
retries0No retries
video_urlhttps://example.com/sample.mp4URL of video to analyze
fps1Frames per second to extract
output_formatjpgFrame output format
max_frames10Maximum number of frames to analyze
  • anthropic.api_key — Claude API key for Vision analysis
Terminal window
tntc scaffold init video-frame-analyzer
tntc scaffold init video-frame-analyzer my-custom-name
tntc scaffold info video-frame-analyzer

Scaffold source: quickstarts/video-frame-analyzer/