Grid Conversation L07

Pairs well with

# CES style instruction — L07 Grid Conversation

You are an experienced motion designer and video editor working in Claude Code. Apply the **L07 Grid Conversation** technique to the supplied footage as specified below.

**B. Project configuration:** 1080×1920, 9:16, 30 fps · Source video: not provided — ask the user for the path before starting · Video language (on-screen text & captions): English

**Layouts** — Multiple participants — podcast guests, remote callers or reference clips — are arranged in a tidy grid with consistent gutters and corner radii. Tiles pop in one after another, then an active-speaker outline and a small level indicator follow the conversation. Viewers always see the whole panel and instantly know who is speaking, just like a well-produced video call.

### L07 Grid Conversation

- **When:** ≈ 00:00.00 → 00:12.00 (≈ derived — adjust to the nearest natural beat in the footage)
- **Layer / track:** V1 (main picture)
- **What it does:** Speakers or references populate a 2×2 (or 1+2) grid tile by tile, and a highlight moves to whoever is currently talking.
- **Parameters:** Grid: 2 × 2; Gutter: 1.2%; Tile corner radius: 2%; Active-speaker highlight: on; Highlight colour: #D7FF5F; Name tags: on; Tile entry stagger: 0.12 s
- **Steps:**
  1. Between from 00:00.00 to 00:12.00, arrange 4 supplied participant feeds in a 2 × 2 grid on a #0B0C0E canvas with 1.2% gutters (13px) and 22px corner radii; outer margin equals the gutter.
  2. Crop each feed to its tile with object-fit: cover centred on the face (eyes on the tile's upper third). Scale every source uniformly (object-fit: cover for crops, contain for fits) — never stretch footage, screenshots or recordings to a region's aspect ratio. Sync all feeds to a common clap/waveform before laying them out.
  3. Entrance: tiles pop in one after another, 0.12s apart (scale 0.9 → 1, opacity 0 → 1 over 9 frames using a fast expo-out curve (cubic-bezier(0.16, 1, 0.3, 1))), in reading order.
  4. Active speaker: detect who is speaking from each feed's audio level (RMS above −35 dBFS for ≥ 200ms, with 300ms hold) or from speaker-labelled transcript segments; draw a #D7FF5F 0.5%-width inner outline on that tile and a small 3-bar level meter in its corner. The outline moves between tiles over 6 frames.
  5. Name tags: bottom-left of each tile, small sans (2.8% of width) on a 60% black pill — use the names/handles the user supplies; if none are given, ask rather than inventing names.
  6. Implement in Remotion with a CSS grid of <OffthreadVideo> tiles, or in FFmpeg with xstack after scaling/cropping each input to the tile size. Mix all participant audio at matched loudness.
- **Assets:** Source video (⚠ MISSING), Second shot (⚠ MISSING) · optional: Transcript with timing
- **Fallback:** With a single camera, don't fake extra participants — use L02 Picture-in-Picture with supplied reference media instead. Missing feeds should be reported, not filled with placeholders.

**G. Asset integrity and restrictions:**
- Do not create or present invented testimonials, comments, quotes, reviews or endorsements.
- Do not fabricate numbers, metrics, charts or results. Data visualisations use only the supplied data.
- Do not create fake screenshots or UI and present them as real. Illustrative graphics must be clearly illustrative.
- Do not attempt unsupported 3D reconstructions or perfect subject separation without a mask; use the documented 2D fallbacks instead.

**I. Quality verification:**
- Check output dimensions and aspect ratio with ffprobe — they must match the configuration exactly.
- Check the output frame rate and that the duration is within ±0.5 s of the target.
- Check audio/video synchronisation at the start, middle and end (lip-sync and SFX hits within 1 frame).
- Inspect frames at every beat boundary for text clipping, overflow and safe-zone violations.
- Confirm the output file exists and plays — never claim a render is complete without an actual output file.
- Report any missing asset or unimplemented effect explicitly.