# Video Production Pipeline

> End-to-end workflow for producing a short video with an agent: brief, script, storyboard and shot list, choosing a renderer (Remotion, HyperFrames, or FFmpeg-only), rendering, and a review checklist before delivery. Hands framework-specific code to the official Remotion and HyperFrames skills. Use this skill when the user wants to plan or produce a video (explainer, launch video, social clip, data animation) and needs structure before any code is written, or needs help choosing a rendering approach. Activate when: make a video, video pipeline, storyboard, shot list, video script, Remotion or HyperFrames, which video framework, render a video, programmatic video, launch video, explainer video.

- Skill: `latestaiagents/video-production-pipeline` (Agent Skill)
- Install (CLI): `npx get-skill add latestaiagents/video-production-pipeline`
- Raw SKILL.md: https://api.skillmd.com/api/skills/latestaiagents/video-production-pipeline/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: latestaiagents (https://skillmd.com/u/latestaiagents)
- Updated: 2026-11-14
- Page: https://skillmd.com/skills/latestaiagents/video-production-pipeline

---


# Video Production Pipeline

**Plan the video before writing a single frame: brief → script → storyboard → renderer choice → render → review.**

Most failed agent-made videos fail upstream: no clear audience, a script that is too long for the runtime, or a renderer chosen before anyone knew what the video needed. This skill is the production plan. Framework-specific code belongs to the official framework skills (see Step 4).

## When to Use

- A user asks for "a video" and the goal, length, or platform is unclear
- Planning a launch video, feature announcement, explainer, or social clip
- Deciding between Remotion, HyperFrames, or plain FFmpeg
- Turning an existing script, blog post, or demo into a shot list
- Reviewing a rendered video before it ships

## When NOT to Use

- Writing Remotion components or HyperFrames compositions: use the official skills (Step 4)
- Editing existing footage only (trim, normalize, captions): use `ffmpeg-post-production`
- Screen-recorded product walkthroughs: start with `product-demo-video`

## Workflow

### Step 1: Write a One-Page Brief

Do not start until these are answered. Ask the user for anything missing.

| Field | Example |
|-------|---------|
| Goal | "Get existing users to try the new export feature" |
| Audience | "Current customers, technical, watching in-app" |
| Platform(s) | YouTube (16:9), Shorts/Reels (9:16) |
| Target length | 30 s |
| Single message | "Exports now run in the background" |
| Call to action | "Open Settings → Export" |
| Assets available | Logo SVG, brand colours, screen recording, VO? music? |
| Sound-off viewing? | Assume yes for social: on-screen text and captions carry the message |
| Deadline / review owner | Who approves the cut |

One video, one message. If the brief lists three messages, propose three videos.

### Step 2: Script to the Clock

Write the script as a two-column table: what is heard, what is seen. Time it.

```markdown
| # | Time | Audio (VO / music / SFX)              | Visual                                  |
|---|------|---------------------------------------|-----------------------------------------|
| 1 | 0–3s | Music in. VO: "Exports used to block…" | Product UI, spinner, title card fades in |
| 2 | 3–9s | VO: "Now they run in the background."  | Click Export → toast appears            |
| 3 | 9–…  | …                                     | …                                       |
```

Timing rules of thumb:

- Read the VO aloud with a timer, or measure a TTS draft. Do not estimate from word counts alone.
- On-screen text needs reading time: hold a line long enough to read it twice at a relaxed pace.
- Leave silence. A 30 s video rarely carries more than a few short sentences of VO.

### Step 3: Storyboard and Shot List

For each row of the script, define the shot so the renderer work becomes mechanical:

```yaml
- shot: 2
  duration_s: 6
  layout: "UI recording, centered, 80% width; caption lower third"
  text_on_screen: "Exports now run in the background"
  motion: "UI slides in 12 frames, ease-out; toast highlight pulse once"
  assets: ["recordings/export.mov", "brand/logo.svg"]
  audio: "VO line 2; music ducked"
  aspect_variants: ["16:9", "9:16 (UI cropped to the toast region)"]
```

Decide aspect ratios now (see `motion-design-principles`). Designing 16:9 and "cropping later" to 9:16 usually loses the important part of the frame.

### Step 4: Choose the Renderer

| Need | Remotion | HyperFrames | FFmpeg only |
|------|----------|-------------|-------------|
| Authoring model | React components; frames driven by `useCurrentFrame()` | HTML/CSS/JS compositions with seekable animations | Filter graphs on existing media |
| Best fit | Teams already in React; data-driven or templated videos; apps that render video for users | Agent-authored motion graphics from plain web tech; editable HTML projects | Trim, concat, reframe, normalize, captions, encode existing footage |
| Rendering | Headless browser + FFmpeg (local CLI, server-side, or Remotion Lambda / Cloud Run) | Headless Chrome seeks each frame, FFmpeg encodes | FFmpeg directly |
| Requirements | Node.js; see Remotion docs | Node.js 22+ and FFmpeg | FFmpeg build with the filters you need |
| Licence | Free for individuals, for-profits with up to 3 employees, non-profits, and evaluation; otherwise a Company License is required | Apache-2.0 | LGPL/GPL depending on build |
| Official agent skill | `npx skills add remotion-dev/skills` | `npx skills add heygen-com/hyperframes`, or `claude plugin marketplace add heygen-com/hyperframes` then `claude plugin install hyperframes@hyperframes` | — (use `ffmpeg-post-production`) |

Decision rules:

1. **No new graphics, only existing footage?** FFmpeg only.
2. **The video is a template filled from data (per-user, per-release, per-metric)?** Remotion fits well, especially if the team ships React; check the licence for commercial teams.
3. **One-off motion graphics authored by an agent, and the team wants plain HTML they can open and edit?** HyperFrames.
4. **Both are viable?** Pick the one the maintainers can read. A video project is code someone will reopen next quarter.

Whichever you choose, load that framework's official skill for code. HyperFrames ships a `/remotion-to-hyperframes` skill if you are migrating; do not hand-translate.

### Step 5: Build in Passes

1. **Animatic:** placeholder boxes and text at final timings, VO or a scratch track. Approve timing before polish.
2. **Layout and type:** final fonts, colours, safe areas, all aspect variants.
3. **Motion:** easing and transitions (see `motion-design-principles`).
4. **Audio:** music bed, SFX, ducking under VO, loudness normalization (see `ffmpeg-post-production`).
5. **Captions:** generated and corrected (see `captions-and-transcription`).

Render low-resolution drafts or single stills for review between passes. Full-quality renders are for the final pass.

### Step 6: Render and Encode Deliverables

- Render a master at the composition's native resolution and frame rate.
- Encode platform deliverables from the master (H.264/AAC MP4 with `+faststart` is a safe web default; recipes in `ffmpeg-post-production`).
- Name files so they sort and identify themselves: `2026-11-export-launch_9x16_v3.mp4`.

### Step 7: Review Checklist

Watch the final file in a real player, start to finish, once with sound and once muted.

- [ ] The single message lands in the first few seconds, with sound off
- [ ] Every text card is on screen long enough to read twice
- [ ] Text and key UI sit inside safe areas in every aspect variant
- [ ] No more than three flashes in any one second (WCAG 2.3.1)
- [ ] Captions present, accurate, and timed; speaker and meaningful sounds noted (WCAG 1.2.2)
- [ ] Spelling of product names, people, numbers checked against the brief
- [ ] Audio: no clipping, VO intelligible over music, loudness normalized to the agreed target
- [ ] First frame works as a poster; a chosen thumbnail exported
- [ ] Licences cleared for music, fonts, footage, and the renderer itself
- [ ] File plays from the start in a browser (moov atom at front)

## Best Practices

1. One video, one message, one call to action
2. Approve the timed animatic before investing in polish
3. Plan every aspect ratio at storyboard time, not at export time
4. Treat the video project as code: version it, keep assets in the repo or a referenced store, script the render
5. Defer framework-specific code to the official Remotion or HyperFrames skill
6. Never describe a render as "done" until someone has watched the exported file end to end
7. Keep a master render; re-encode deliverables from it rather than from other deliverables

## References

- Remotion agent skills: https://github.com/remotion-dev/skills and https://www.remotion.dev/docs/ai/skills
- Remotion licence: https://github.com/remotion-dev/remotion/blob/main/LICENSE.md
- Remotion render CLI: https://www.remotion.dev/docs/cli/render
- HyperFrames (README, skills, requirements, licence): https://github.com/heygen-com/hyperframes
- HyperFrames docs: https://hyperframes.heygen.com/introduction
- WCAG 2.2 SC 1.2.2 Captions (Prerecorded): https://www.w3.org/WAI/WCAG22/Understanding/captions-prerecorded.html
- WCAG 2.2 SC 2.3.1 Three Flashes or Below Threshold: https://www.w3.org/WAI/WCAG22/Understanding/three-flashes-or-below-threshold.html
- YouTube recommended upload encoding settings: https://support.google.com/youtube/answer/1722171
- FFmpeg licence: https://ffmpeg.org/legal.html

