---
name: editorial-short-form-video-production
description: Create, redesign, render, and quality-check premium short-form vertical video (Instagram Reels, TikTok, Shorts), especially from horizontal source footage. Use when the task requires consistent typography, safe zones, face-preserving reframing, branded editorial composition, FFmpeg rendering, covers, or release QC—not just scripts or captions.
version: 1.0.0
author: Hermes Agent
license: MIT
metadata:
  hermes:
    tags: [reels, short-form-video, motion-design, ffmpeg, editorial-design, quality-control]
---

# Editorial Short-Form Video Production

## Purpose

Produce finished, premium vertical videos—not merely concepts. Treat a multi-Reel batch as one editorial collection with a shared typographic, compositional, color, transition, and branding system.

## Production sequence

1. Inspect every source technically with `ffprobe`.
2. Build dense timestamped contact sheets for long footage.
3. Select clips based on complete motion ranges, not favorable stills.
4. Write one thesis, one turn, and one final line per Reel.
5. Define a series-wide design system before rendering.
6. Build vertical compositions without sacrificing faces or gestures.
7. Render all videos and covers.
8. Perform visual and technical release QC.
9. Recheck automated visual flags against the actual frame/video before re-editing.
10. Deliver individual MP4s plus one archive containing covers and posting notes.

## Human-framing integrity

Face integrity outranks filling the entire 9:16 canvas.

- Whenever a face is visible, preserve the forehead, chin, both eyes, and hair contour for the entire clip range.
- Inspect temporal motion; a safe still does not make the whole range safe.
- If the source already cuts the face in-camera, exclude that range. Showing the full source frame cannot repair an existing crop.
- If a 9:16 crop cannot preserve the person, retain the full source frame in an editorial window within the vertical canvas.
- If that still fails, choose full body, silhouette/back view, landscape, hands, or process footage. Never use an accidental fragment of a face.
- For intentionally full-body compositions, preserve the head, hands, feet, and meaningful gestures.

## Typography

- Use one actual font family throughout a series; weight/style variation inside that family is acceptable.
- Do not use native-app-looking grey rounded caption boxes unless explicitly requested.
- Use semantic line breaks; usually no more than two lines per beat.
- Place text in negative space, never across eyes, mouth, face, or meaningful body action.
- Verify the renderer used the intended physical font. Merely naming the family in ASS/SRT styling is insufficient because libass/fontconfig may silently substitute a system font.

## Composition and safe zones

For 1080×1920 vertical video, use conservative working margins unless current platform previews prove otherwise:

- top of key text: below y=280;
- bottom of primary message: above y=1380;
- side margins: at least 130 px;
- reserve the bottom 360 px and rightmost 150 px for platform UI;
- keep branding above bottom UI;
- ensure cover text also survives profile-grid crops.

Safe zones vary with placement and platform UI. Always preview the final asset in the intended destination when possible.

## Visual rhythm

- Make the hook legible in the first 0.3–0.5 seconds.
- Introduce a visual or textual beat every 1.5–2.8 seconds.
- Keep the final statement readable for at least 2.2 seconds.
- Motivate transitions through meaning rather than decoration.
- Avoid effects that resemble missing-media boxes, stickers, or accidental UI.
- Keep CTA in the caption for contemplative text-led work unless an on-video CTA is explicitly required.

## Rendering standard

Default delivery:

- 1080×1920, 9:16;
- H.264 High Profile;
- yuv420p;
- 30 fps;
- `+faststart`;
- typically 11–16 seconds for text-led Reels;
- no commercial music embedded when a current native platform sound will be added at publication.

## Release gate

Do not call a Reel finished until all apply:

- [ ] Every visible face is complete across the full selected interval.
- [ ] Full-body gestures are not accidentally clipped.
- [ ] One verified font family is used consistently.
- [ ] Text is readable, semantically broken, and inside safe zones.
- [ ] Text does not cover people or meaningful action.
- [ ] Branding remains clear of native UI.
- [ ] Foreground video is present throughout.
- [ ] No unintended blank or black frames exist.
- [ ] The final statement has sufficient reading time.
- [ ] The file decodes end-to-end without errors.
- [ ] Resolution, fps, codec, and pixel format match delivery requirements.

## Automated-review caution

Contact-sheet and vision-model QC is useful for finding suspected defects, not proving them. Artwork, laptop screens, intentional dark fields, and subtitle fades can be misclassified as UI boxes, missing media, or absent text. Validate every flag against the exact timestamp, subtitle event file, source footage, and technical probes before re-editing.

## References

- `references/cosmos-luminosity-reels.md` — Cosmos Luminosity-specific editorial system, palette, strict face rule, FFmpeg font verification, and QC workflow learned from a five-Reel production batch.