LlamaCasty/TASKS/task-53-output-formats.md

TASK 53 — Output-format verification: vertical and landscape outputs (queued 2026-10-08)

Catalog: TASKS.md — status and requirements live here.

Why: the site claims vertical 1080p60 (9:16), but the full route from compositor crop through FFmpeg scaling to YouTube and the local recording needs verification.

Goal: prove each output tier has the expected dimensions and frame rate, and that vertical live and recorded outputs preserve the composed image. Creators control layer placement; this task does not promise platform-specific safe zones or keep content clear of every viewer overlay.

Slices (one integration test each, in order):

1. ✅ What we actually emit (2026-11-04; creator visually reviewed and accepted) — six creator-supplied recordings confirm the emitted configurations. All are H.264 with exact average frame rates and dimensions:

   Mode	File	Output
   Landscape 1080p60	rec1-1080p60.mp4	1920×1080 @60 fps
   Landscape 1080p30	rec2-1080p30.mp4	1920×1080 @30 fps
   Landscape 720p60	rec3-720p60.mp4	1280×720 @60 fps
   Landscape 720p30	rec4-720p30.mp4	1280×720 @30 fps
   Vertical 1080p30	rec5-1080p30-vert.mp4	1080×1920 @30 fps
   Vertical 1080p60	rec6-1080p60-vert.mp4	1080×1920 @60 fps

2. ✅ 9:16 output resolution (resolution-only acceptance, 2026-11-04) — creator-supplied ty-20261104-0927-982-0000-2.mp4 reports H.264 1080×1920 at 60 fps. The creator accepted the resolution only; no platform-specific safe-zone promise is part of acceptance.
3. ✅ Vertical YouTube live-player check (creator confirmed 2026-11-04) — the 9:16 stream displays vertically without distortion or letterboxing. Layer placement relative to YouTube controls remains creator-controlled; platform-specific safe zones are not a requirement.
4. ☐ Recorded .mp4 matches output — confirm the vertical local recording uses the same crop and composed picture as the live stream, not merely the same dimensions.

Design decisions:

• Geometry is a contract with tests, not a hope. Derived ratios (stream vs composite vs canvas) get a derived-value test so one change can't silently wreck Shorts/TikTok output.
• Keep widescreen untouched — only add what vertical needs (the existing test suite stays green for 16:9).