research/reference-format-analysis

Reference Format Analysis — extract format DNA from a short-form video.

Category: research
Source: workflows/research/reference_format_analysis.py

Input Schema

Field	Type	Default	Description
`artifact_dir`	`object`	—	Override output directory.
`enable_ocr`	`boolean`	`false`	Run OCR on sampled frames.
`enable_tone_inference`	`boolean`	`true`	Use LLM for tone/hook/archetype inference.
`extract_frames`	`boolean`	`false`	Sample keyframes to disk.
`frame_sample_fps`	`number`	`1.0`	FPS for frame sampling.
`max_duration_sec`	`object`	—	Reject videos longer than this.
`platform`	`string`	`"generic"`	Source platform: instagram_reels, tiktok, youtube_shorts, generic.
`reference_video_path`	`string`	`""`	Local path to reference video file (MP4/MOV).
`reference_video_url`	`string`	`""`	URL to download reference video from (YouTube, TikTok, Instagram, or direct link).
`regenerate`	`object`	—	When set, this run is a regeneration. Workflows may read `direction` / `keep` / `extra_instructions` to modulate prompts; the engine persists `parent_run_id` and `parent_variant_index` as run lineage columns.
`variants`	`integer`	`1`	Number of independent variant executions (1–10). When > 1, the engine runs the workflow N times with different sampling, producing N outputs.
`whisper_model`	`string`	`"large-v3"`	Whisper model size for transcription.

Output Schema

Field	Type	Default	Description
`format_dna_json_path`	`string`	`""`	Path to format_dna.json
`frames_dir`	`object`	—	Path to extracted frames
`kind`	`object`	—	Variant card shape: video / carousel / image / text. Surfaced on the per-variant entry of the run-output API and used by gallery UIs to pick the right layout.
`manifest_json_path`	`string`	`""`	Path to analysis_manifest.json
`ocr_json_path`	`object`	—	Path to ocr.json
`recreation_blueprint_json_path`	`string`	`""`	Path to recreation_blueprint.json
`shots_json_path`	`string`	`""`	Path to shots.json
`summary_markdown_path`	`string`	`""`	Path to summary.md
`transcript_json_path`	`string`	`""`	Path to transcript.json

Task Pipeline

validate_and_probe → detect_shots_and_visual_analysis → transcribe_and_analyze_audio → infer_format_dna → emit_analysis_artifacts

Task	Description
`validate_and_probe`	Download (if URL provided), validate input video, and extract metadata.
`detect_shots_and_visual_analysis`	Detect shots, classify composition, analyze motion and color.
`transcribe_and_analyze_audio`	Transcribe speech, map pauses, detect retention hooks, analyze audio layers.
`infer_format_dna`	Use LLM to infer structural patterns, then assemble FormatDNA.
`emit_analysis_artifacts`	Write all analysis artifacts to disk.

Run-spec example

Save the YAML below as my-run.yaml, edit the values, and run with the CLI or POST it to the API. Required fields are uncommented; optional knobs are documented above the input: block — copy any line under input: and uncomment to set.

workflow: research/reference-format-analysis

# Optional fields — copy any line(s) under `input:` and uncomment to set:
# Override output directory.
#   artifact_dir: null
#
# Run OCR on sampled frames.
#   enable_ocr: false
#
# Use LLM for tone/hook/archetype inference.
#   enable_tone_inference: true
#
# Sample keyframes to disk.
#   extract_frames: false
#
# FPS for frame sampling.
#   frame_sample_fps: 1.0
#
# Reject videos longer than this.
#   max_duration_sec: null
#
# Source platform: instagram_reels, tiktok, youtube_shorts, generic.
#   platform: generic
#
# Local path to reference video file (MP4/MOV).
#   reference_video_path: ""
#
# URL to download reference video from (YouTube, TikTok, Instagram, or direct link).
#   reference_video_url: ""
#
# Whisper model size for transcription.
#   whisper_model: large-v3
#

input: {}

Run it locally:

fab-workflow --from-file my-run.yaml

Or submit over the wire — the same file is the request body:

curl -X POST 'https://gofabric.dev/v1/workflows/runs?name=research/reference-format-analysis' \
  -H 'Authorization: Bearer fab_xxx' \
  -H 'content-type: application/yaml' \
  --data-binary @my-run.yaml

Every workflow also accepts the universal WorkflowInput fields — variants (1–10 fan-out) and regenerate (creative-direction hints with run lineage). See Run-specs (YAML / TOML / JSON) for the full top-level shape (metadata, priority, bundle, parent, etc.).