Quiet cinematic 2D motion-comic short about floating numbers over people

[Note: the creator shared a partial prompt. Wording below is theirs.] GOAL: <inputs> Ask me for my story idea. MOTION: It's a "What if you could see ___ above everyone's head" premise plus a ladder of 3-4 people, from a stranger to someone who really matters. Or pitch 3 of them (e.g. times left you'll see someone, minutes left you'll talk to someone, how many more texts you'll send someone). Ask what my @handle is. Ask which voice tool I have: free first (Kokoro or Piper, run locally), or ElevenLabs (about $0.45 for a 2-minute Short). Ask if I have an image or video model, and tell me this method needs neither, because every frame is drawn in code. Then write the script and a shot list with seconds, and show me both before anything costs money. </inputs> <direction> A 1080x1920, 30 fps, 2:00-2:15 vertical Short in the style of the "Times Left" animated story shorts: a quiet, cinematic 2D vector motion comic, drawn entirely on an HTML canvas in code. Not 3D, not AI images. The tone is emotional, not hyper: it should feel like a short film, with one gut-punch in the middle and a hopeful ending. Characters: one blob-stick rig shared by the whole cast. - At scale 1: a perfectly round solid-black head, 150 px across, with one curled tuft on top (a 40 px teardrop with a curl, like a sprout). The body is an oversized jacket shape, 130x220 px, with rounded shoulders and thick sleeve-arms (34 px wide) ending in round nub hands, no fingers. Legs are 10 px black lines, 200 px long, with 34x14 oval feet. No outlines on anything. Each jacket has one flat darker shade down one side and a lighter zipper or seam line. - The face is drawn in WHITE on the black head: eye whites as ovals 26x30 with 10 px black pupils, brows as 6 px white strokes, mouth as a 6 px white stroke. No nose, no ears. - Draw at scale 1.0 in wide shots, 1.6-2.2 in medium shots, and 3.5-4.5 in close-ups, where the head fills about 60% of the width and the shoulders are cropped by the bottom of the frame. - One look each. The hero ("you"): an off-white puffer hoodie #e9e6df with a grey zip #b9b5ad and the hood bunched at the neck. The person who matters (best friend or sibling): a navy-blue jacket #2f6bb0 and a red cap #d8413a worn backwards. Side characters get one jacket colour each: a green apron #4c9a6e (barista), a mauve coat #9a5c8c (commuter), a burnt-orange coat #d8662e (the stranger at the end, holding a blue book #4a7fd0). - Faces carry the feeling. Eyes: normal, wide (shocked), happy arcs, half-closed, narrow (thinking), one-tear (a blue drop #4fa3ff rolling down). Brows: up, worry, angry, flat, one raised. Mouths: smile, grin, o, flat, wavy, frown, open laugh. Blink for 0.1 s about every 2.5 s. Heads tilt 3-6 degrees when someone talks. The floating number (the whole hook): - Above EVERY head, about 50 px over the tuft: a glowing cyan dot-matrix LED number, 110 px tall at scale 1, built from round dots on a 5x7 grid. Core #bff6ff, glow #2fc6ff with a 30 px bloom, and a subtle flicker (opacity 0.92-1, about 8 Hz, random). - When a number changes, it rolls like a slot machine: 0.35 s of vertical motion blur, overshoot, then settle with a small camera shake. Big jumps (3 -> 26 -> 515 -> 27,480 -> 60,000) roll faster each time. - In close-ups the number can be cropped by the top of the frame, but at least half of it stays visible. Sets: soft flat vector places with gentle gradients, no outlines, a muted cinematic palette, and depth of field (the background blurred 4-10 px, the foreground sharp). A light film grain (3% noise) and a soft vignette over everything. - Train interior: pale green-grey #c9d3cf, grab handles, ad strips in red/pink/green, blue seats #33539a. - Café: warm beige #e8d6b8, a shelf of colourful cups, a wooden counter, a pendant lamp. - Bar: dark brown #3a2418, a back wall of glowing bottles (amber, green, blue, red), a red leather booth, string lights. - Night: deep blue #1c2236 for phone moments. - Airport: cool white-blue #dfe8f0, big windows, a plane outside, navy queue ropes. - Bus stop: pale sky #b9d0e0, a glass shelter, a red poster ad, beige buildings behind. Motion: slow and filmic, but never frozen. - The camera pushes in 4-10% across every shot, with an ease-in-out. A small shake (decaying over 0.3 s) only when a number changes or something lands emotionally. - Cuts every 2-5 s. Alternate wide shots (the number above everyone), medium two-shots and face close-ups. - Characters always have life: breathing (scale y 1.0-1.015 over 3 s), blinks, small head turns. Walks are a simple 2-frame leg swing with a 6 px bob. - Things come in with a soft slide plus a scale from 0.96, never a hard pop. Props (paper, phone) slide up from the bottom with a slight rotation (-3 to 3 degrees). Overlay UI (screen space, always on top): - Handle: "@YOURHANDLE", 30 px, white, all caps, letter-spaced, top-left at x 44, y 136. - Header, 2 lines, centred at x 540: line 1 is the person's label ("EVERY STRANGER", "THE BARISTA", "YOUR BEST FRIEND", "BUS STOP GIRL"), white, at y 262. Line 2 is "TIMES LEFT: X" in yellow #ffc83d at y 366. Both are Montserrat Black at 78 px, all caps, with a 12 px black outline and a soft drop shadow. The header always matches the number of the person in focus, and it ticks at the same moment that number rolls. - Date pills: small white rounded pills with a black outline, 26 px letter-spaced caps ("FRIDAY NIGHT", "WEEK 7 · SAT", "WEEK 8 · TUE"), centred at y 490. They slide in and fade out after 2 s. - Props drawn as world objects: a torn-edge paper checklist with empty/ticked checkboxes and a highlighter swipe, a phone with chat bubbles, a phone showing a group photo where one person is a dotted empty outline, a flight-booking app with a yellow BUY button, a red handwritten "= 2 ?". - End pill: "FULL FILM ON THE CHANNEL" at y 1530 for the last 3 s. Captions: - 1-3 words at a time in Montserrat Black, 92 px, centred at y 1255, max 940 px wide, all caps. - White with a 14 px black outline and a soft shadow, and ONE key word per chunk in yellow #ffc83d (the noun or verb that matters). - Each chunk pops from 85% to 100% in 0.12 s. Keep full stops, commas and question marks as in the original style ("THAT'S NEW.", "COFFEE?"). - Character dialogue gets the same captions; no speaker labels. Banned: the @otherwise handle or any real creator's handle, real logos or airline/app brands, AI stills, a frame that holds completely still for more than 1.5 s, world text on the caption line (y 1190-1320), bright saturated "kid" colours, whispered narration. </direction> <structure> Length: 2:00-2:15, 280-330 words. Times below come from a 133 s reference; scale them to the voice. 0-6 s: the HOOK. The narrator states the premise as line 1 ("Every person on this train has a number floating over their head") over the most striking picture: a wide shot where every head shows the same number. Push in slowly. 6-12 s: the rule clicks. A close-up of your face going wide-eyed, then one short reaction line ("Okay. That's new."). 12-19 s: the LIGHT BEAT. A side character with a big number (212). One small exchange, and the number ticks down by one (211). A tiny laugh. 19-30 s: the CORE PERSON enters with a small number (3). One line that shows how much they matter ("your best friend since you were eight"). 30-40 s: the COUNTDOWN. They list upcoming plans; a checklist shows the plans, and the count of plans equals the number. You say "Wouldn't miss it." 40-50 s: you SKIP one. A phone text ("Next time, I promise"). The number drops. A line from them that stings ("You say that a lot"). 50-62 s: the GUT-PUNCH. The checklist math doesn't add up ("= 2 ?"), then the group photo with you as an empty outline: "The party happened. You just weren't in it." 62-75 s: the GOODBYE. The last event on the list, a hug, one tear, number at 1. One quiet line from them ("Did I do something?") and yours ("No. I did."). 75-97 s: the TURN. "And then you have a very stupid thought": if skipping can take a number away, showing up can add one. You book something, and the number climbs 1 -> 2 -> 3 -> 41 with them cheering, arms up. 97-129 s: the CALLBACK. Back to a stranger you see every day, number 1. "You could keep her on pause forever. Or you could find out what comes after one." You talk. Their line reveals they were waiting too. The number rockets 3 -> 26 -> 515 -> 27,480 -> 60,000 over a short time skip. 129-133 s: the CLOSING LINE, the theme in one sentence, over the final wide shot plus the end pill. It can end mid-thought to drive replays. </structure> <build> 1. Tools (free): Python 3, Playwright with Edge or Chromium, FFmpeg, faster-whisper, numpy, Pillow. Install what's missing and tell me what you installed. 2. Script (free): - Point of view: "you" and "your" for the hero. The narrator is calm and warm, like a friend telling you a story. Dialogue lines are short and real, how people actually talk. - Every beat is about the number: it appears, it ticks, it drops, it climbs. If a beat doesn't move a number or make you feel something about one, cut it. - One picture per line, 6-16 words per line. - The light beat is a breather, the core person carries the story, the stranger is the payoff. The number must drop once (the cost) and climb twice (the fix, then the dream). 3. Voice: record one take per character, with all of that character's lines separated by blank lines. - ElevenLabs eleven_v4 with stock library voices, never a cloned real creator. Narrator: a calm young man, warm and slightly amused (median pitch 110-150 Hz). Hero: a natural young guy. Best friend: a louder, joking guy. Stranger: a soft, playful young woman. - Use audio tags sparingly: [softly], [laughing], [sad], [excited]. Cost: about 2,300 credits, about $0.45 for a 2:10 Short. - Free route: Kokoro or Piper locally. Pick the warmest voices; it will sound flatter. - Cut the takes into lines with faster-whisper word times (pad 0.06 s before and 0.2 s after). Keep natural pauses inside a line up to 0.35 s; this style breathes. Leave 0.3 s between lines, 0.6-0.9 s before the gut-punch and the turn. - Match each take's level to the narrator, then lift any line more than 1.5 dB under the narrator's median. Use atempo only up to 1.05. - Run whisper again on the finished track for word times. Every cut, number tick, caption and sound starts on a real word. 4. Film (free): one HTML page with one canvas, where every frame is a pure function of time: https://t.co/KjSvWkCgNW(t), window.READY, window.END. - A SHOTS list of [start, end, function]. Each shot sets a camera (centre x, centre y, zoom, keyed over time), draws its world, and may return an overlay drawn in screen space. - One rig function draws every character from a pose object: x, y, scale, breathe, rotation, arm and leg angles, head tilt, eyes, brows, mouth, look, tear, number value, number roll progress. - Find hand positions with the same arm maths, so held props (a phone, a coffee cup, a beer, a book) sit in the hand. - A STATE timeline per person ([time, value]) drives both the floating number and the header, so they can never disagree. - Montserrat Black (free on Google Fonts) for every word. 5. Check before render: draw 3 contact sheets of 12-16 frames at key times. Fix captions over world text, numbers cut off, the header not matching the number, characters cut off and dead empty space. 6. Render (free): seek every frame in headless Edge or Chromium and encode inside the page with WebCodecs H.264 on the GPU: 30 fps, 40 Mbps, 2 motion-blur subframes over half a frame. - About 4,000 frames; expect 4-6 minutes. - Re-encode with libx264 CRF 20, preset slow, for the upload. 7. Sound (free): - Music: one soft piano or lo-fi bed (Mixkit, free licence). Drop it to silence for the gut-punch line and on "No. I did.", bring it back swelling on the turn, and let it peak on 60,000. - Effects: Mixkit, placed so each effect's loudest moment lands on its frame. Train rumble and café/bar room tone as low beds, a soft digital tick on every number change, a whoosh on cuts between places, a phone buzz, a paper slide, a pen scribble on "= 2 ?", a plane pass at the airport, a bus pass, a soft chime on the final number. - Keep the narrator 15+ dB over everything else on every line. Render a voice-only mix and a no-voice mix with the same gains (before loudness normalising), compare RMS per line, and fix any line under 15 dB. - Starting gains: music bed about 20 dB down, room tones 24 dB down, effects on speech 12 dB down, effects in gaps 5 dB down, every cue capped at -3 dBFS peak. - Two-pass loudnorm to -14 LUFS, true peak -1.5. 8. Cover (free): - Study the top 5-10 covers in t