Clip Workflows

Clip workflows transform longer videos into short-form viral content. They handle moment detection, clip extraction, reframing, subtitles, hooks, dubbing, and effects.

Clip Generator

Workflow: clip/clip_generator

Complete pipeline: download a video, detect viral moments, extract clips, reframe to vertical, add subtitles, hooks, and effects.

# From YouTube
fabric run clip/clip_generator --input url="https://youtube.com/watch?v=..."

# From local file
fabric run clip/clip_generator --input media_path="/path/to/podcast.mp4"

Pipeline

transcription_pipeline
        │
  detect_viral_moments     (Gemini moment scoring)
        │
  extract_clips            (FFmpeg at timestamps)
        │
  reframe_pipeline         (MediaPipe + YOLO → 9:16)
        │
  subtitle_pipeline        (word-level subtitles)
        │
  hook_overlay_pipeline    (AI text overlays)
        │
  effects_pipeline         (AI color grading)
        │
  collect_final_clips

Input

Parameter	Type	Default	Description
`url`	`string`	—	YouTube or video URL
`media_path`	`string`	—	Local video file
`query` / `topic`	`string`	`""`	Focus topic for moment detection

Output

{
  "clips": [
    {
      "id": "clip_001",
      "path": "/tmp/clip_001_final.mp4",
      "hook": "You won't believe what happened next",
      "start": 45.2,
      "end": 75.8,
      "duration": 30.6,
      "score": 0.92
    }
  ],
  "total_clips": 5
}

Blur Reframe

Workflow: clip/blur_reframe

Converts landscape to 9:16 vertical without cropping — the full frame is visible with a blurred background fill (pillarbox style).

fabric run clip/blur_reframe --input media_path="/path/to/landscape.mp4"

Output

{
  "vertical_path": "/tmp/blur_reframe.mp4",
  "reframe_mode": "blur"
}

AI Dubbing

Workflow: clip/dubbing

Translates and dubs video audio into 30+ languages using ElevenLabs voice synthesis.

fabric run clip/dubbing \
  --input media_path="/path/to/video.mp4" \
  --input target_language="Spanish" \
  --input full_text="Original transcript..."

Pipeline

translate_transcript → generate_dubbed_audio → replace_audio_track

Input

Parameter	Type	Default	Description
`full_text`	`string`	required	Original transcript
`target_language`	`string`	required	Target language name
`voice_id`	`string`	auto	ElevenLabs voice ID
`elevenlabs_api_key`	`string`	env	API key

Output

{
  "dubbed_path": "/tmp/dubbed_spanish.mp4",
  "dubbed_audio_path": "/tmp/dubbed_audio.mp3"
}

Hook Overlays

Workflow: clip/hooks

Generates attention-grabbing text and burns it as an animated overlay on the first 3 seconds of a clip.

fabric run clip/hooks \
  --input media_path="/path/to/clip.mp4" \
  --input hook="This changes everything"

Pipeline

generate_hook_text → burn_hook_overlay

If no hook text is provided, the LLM generates one from the transcript excerpt.

Output

{
  "hooked_path": "/tmp/hooked_clip.mp4",
  "hook_text": "This changes everything"
}

Video Effects

Workflow: clip/effects

Uses Gemini to generate professional FFmpeg filter chains based on content type and mood, then applies them.

fabric run clip/effects \
  --input media_path="/path/to/clip.mp4" \
  --input mood="dramatic" \
  --input content_type="talking_head"

Pipeline

generate_effect_filters → apply_effects

Output

{
  "effects_path": "/tmp/effected_clip.mp4",
  "ffmpeg_filters": "eq=contrast=1.1:brightness=0.02,unsharp=5:5:0.5"
}