faceless-mixed-short

name	faceless-mixed-short
description	Build a faceless short that mixes voiceover, stock footage, local footage, gameplay, AI-generated images or clips, captions, and music into one coherent edit.

Faceless Mixed Short

Use When

• The user wants a net-new short without a human presenter on camera.
• The strongest output will come from mixing visual sources, not forcing one source for every beat.
• The short needs voiceover, captions, music, and pattern interrupts.

Core Approach

1. Write for voiceover-first pacing.
2. Use the simplest source that sells each beat: stock, local footage, gameplay, generated image, generated clip, or graphic card.
3. Mix sources deliberately. Variety should create rhythm, not noise.
4. Use captions as part of the composition, not as an accessibility afterthought.
5. Reject outputs where the visual layer looks like keyword wallpaper.

Inputs

• topic, angle, or packaging promise
• optional brand/niche hints
• optional local assets or gameplay
• provider preferences for stock or generated media

Outputs

• script.json
• audio.wav and timestamps
• visuals.json
• final MP4 plus review bundle

Proven Example

• Local proving bundle: experiments/proving-wave-3/faceless-mixed-short/outputs/final/video.mp4
• Tracked preview: docs/demo/demo-15-faceless-mixed-short.mp4
• Pattern used: practical tech hook, cold-phone opener, simplified chemistry diagram, phone UI recovery card, three practical tip cards, recap card, CTA, burned-in captions, quiet synthesized music bed, and final loudness normalization.
• Publish-prep passed portrait resolution, duration, format, cadence, and audio-signal checks. OCR caption-sync still needs a proper caption export sidecar for FFmpeg fallback renders.

Optional Runtime Surface

• Compose this with brief-to-script, script-to-audio, timestamps-to-visuals, video-render, and publish-prep-review.
• Use short-form-production-playbook as the editorial layer.

Technical Notes

• Gameplay and stock are support footage, not excuses to avoid visual judgment.
• AI-generated visuals work best as emphasis beats, impossible shots, diagrams, or stylized transitions.
• Use music and motion changes to separate beats, not to hide weak writing.
• Route music beds, SFX, ambience, and YouTube-origin audio through audio-source-policy before direct use in public demos.
• Do not place secondary explanatory text behind captions. The main visual should carry context, and captions should occupy the clean caption zone without text collisions.
• Mixed-source does not mean random-source. Every source change should map to a beat change: hook, mechanism, consequence, proof, tip, recap.
• If using FFmpeg fallback assembly, still include a real voiceover, audible but quiet music bed, portrait duration, cadence flashes or real cuts, and publish-prep review on the final MP4.

Aggregated From

• RayVentura/ShortGPT
• rushindrasinha/youtube-shorts-pipeline
• mutonby/openshorts
• this repo's stock, gameplay, NanoBanana, Ken Burns, and Veo paths

Validation Checklist

• The hook lands immediately with sound on or off.
• Visual sources vary on purpose and do not feel randomly swapped.
• Captions remain readable across all source types.
• Music supports pace without masking narration.
• The render feels like one edit, not several pipelines stitched together.
• Contact-sheet review shows no source-text collisions, clipped UI, unreadable captions, or accidental blank/default backgrounds.