# GUIDE.md: "Galloping Gertie" short ## What's here - `short.mp4`: the film, 1080x1920, 30 fps, 59.0 s, H.264 plus AAC, -14 LUFS. - `poster.png`: the hook frame (f41). - `SCRIPT.md`: the facts with sources, then the script. - `STYLE_BIBLE.md`: the look, as numbers. - `ROUNDS.md`: every look round and score; the sheets are in `rounds/`. - `CHECKS.md`: every QA result. - `LESSONS.md`: defect classes to check from the start next time. - `source/`: everything needed to rebuild. There are no reference shorts, so there are no topic reports (`ref/reports/`). The look is the chosen option B, written straight into the bible. ## Rendering Remotion renders in headless Chrome, so run the render on a machine where that's allowed, from the project root (`source/`): ```sh npm ci --no-audit --no-fund mkdir -p .tmp && rm -rf .tmp/remotion-* # Remotion copies public/ into the temp dir on every bundle TMPDIR="$PWD/.tmp" node scripts/render_stills.mjs r7 scripts/stills.json # half-size look stills (about 1-2.5 min for 37) TMPDIR="$PWD/.tmp" node scripts/render_final.mjs # film (silent) plus poster: about 34 min at 4 tabs on the render machine ``` Then, on any machine with ffmpeg and Python (no browser needed): ```sh python3 -m venv .venv && .venv/bin/pip install numpy scipy pyloudnorm faster-whisper opencv-python-headless pillow ./scripts/mux.sh # adds audio/mix.wav as the only audio track -> out/short.mp4 .venv/bin/python -I scripts/qa.py out/short.mp4 # black/white/frozen/pop/length/loudness checks -> out/qa.json ``` - Use `CONCURRENCY=2` if the GPU runs out of memory, and `REMOTION_GL=swangle` if ANGLE fails. - Delete `.tmp/remotion-*` when nothing is rendering. - For 1440x2560, add a composition in `src/Root.tsx`. Every pixel size scales with height. ## Where to change things | Change | Where | |---|---| | Story beats, shot list, camera keys, cuts, crossfades, shakes, blur, light per shot | `src/story.ts` (`SHOTS`) | | Caption text and timing | `src/story.ts` (`CAPTIONS`, `END_LINE`) | | Voice | Replace `audio/voice.wav`. Then run `scripts/transcribe.py` and `scripts/fix_words.py`, which writes `src/words.json`. Then re-time `story.ts`, re-export the events (`scripts/export_events.ts`) and re-make the sound. | | Colours, light presets, sky, fog, toon ramp, painted textures | `src/materials.ts` (`PAL`, `LIGHTS`) | | Bridge (deck motion, twist, collapse), new bridge, splash | `src/world/bridge.ts` | | Terrain, water, sky, clouds, trees, culler, car | `src/world/env.ts`, `src/world/terrain-fn.ts` | | Wind tunnel, models, streamlines | `src/world/tunnel.ts` | | Characters (kid, close-up, Tubby, engineer), faces, blinks | `src/characters.ts` | | Outlines, lens blur, bloom, grade, crossfade | `src/post.ts` | | Camera interpolation and impact shake | `src/camera.ts` | | Music, effects, ducking, master | `sound/make_sound.py`. Re-make the sound after any timing change. | | Framing check before rendering | `npx esbuild scripts/framing_check.ts --bundle --platform=node --format=esm --outfile=.tmp/f.mjs && node .tmp/f.mjs S5` | ## Time it took on the render machine - Half-size look stills: 52-159 s for 34-37 stills. - Final film at 1080x1920 with 4 tabs: 2016 s (33.6 min). - Mux plus QA here: under 1 min.