ProductcommunityClaude Opus (creator-reported)

What If Roblox Story Short

A 45-second vertical "What if" Short in the 2D kid-Roblox story style. Every frame is drawn in code on an HTML canvas, with a shared blocky character rig, punchy captions, fake game UI, voices and a rewind loop back to the first frame. No image or video model.

Remix

How it was made

No additional details were supplied.

Models

  • Claude Opus (creator-reported) · Anthropic · execution

About

A 45-second vertical "What if" Short in the 2D kid-Roblox story style. Every frame is drawn in code on an HTML canvas, with a shared blocky character rig, punchy captions, fake game UI, voices and a rewind loop back to the first frame. No image or video model.

Try it

Copy the prompt and run it with Claude Opus in a coding agent with a shell. It asks for your story idea (or pitches three) and shows you the script and shot list before any voice is generated. Everything is free except an optional ElevenLabs voice, at about US$0.15 a Short. The preview is the creator's actual output, not a Possible-generated substitute.

Creator and prompt

By Finn. See the original post. The copyable text is the prompt as posted, from <inputs> to </start>, with only the post's introduction removed.

What to know

The creator says he recreated the format with Claude Opus for $0, in full code. We haven't run the prompt ourselves or inspected his source. Roblox is a trademark of its owner; this result isn't affiliated with or endorsed by Roblox, and the prompt bans real logos and real creators' characters.

Model provenance: Claude Opus is reported by the creator, with no version given. Possible verified the source pairing and media presentation, not an independent generation. The preview is a re-encoded copy of the creator's video, smaller for playback. Creator media is excluded from the repository's MIT license; the original post remains linked.

Exact prompt

Hidden characters, if present, appear as Unicode escapes
<inputs>
Ask me for my story idea: a "What if" that crosses a kid's real-life pain (screen time, Mom, chores, school, the dentist, a little brother) with one Roblox rule. Or pitch 3 different ones. Ask which voice tool I have: free first (Kokoro or Piper, run locally), or ElevenLabs (about $0.15 a 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, 60 fps, 40-45 s vertical Short in the style of the big 2D "What if" kid Roblox story channels: flat 2D cartoon cut-outs animated like motion graphics, drawn entirely on an HTML canvas in code. Not 3D, not AI images.
Characters: one blocky game-avatar rig shared by the whole cast.
- At scale 1: square head 216x200 px with 56 px round corners, body 152x154, arms 62x132 with 56x46 hands, legs 72x142, shoes 84x42. Draw them at scale 1.0-2.4.
- Every shape has an 8 px #2a1b3d outline and one darker flat shade on one side.
- One look each. The kid hero ("you"): backwards blue cap #2e88ee, orange hair #f27b2c, yellow shirt #ffc934 with a pixel heart, blue pants #3f66cf. Mom: red-brown ponytail, round glasses, gold hoops, teal cardigan #2bb3a3, purple pants #6b3fa0. One or two side characters (a dentist, a teacher, a little brother).
- Faces carry the comedy. Eyes: normal, happy arcs, huge, star, spiral, closed, narrow. Brows: up, worry, angry, flat. Mouths: smile, grin, o, scream, wavy, flat, smirk, tongue, chew, wide. Plus blush, sweat drops and sunglasses that drop in. Blink for 0.1 s about every 1.6 s.
Sets: bright flat places (bedroom, classroom, dentist, a boss office, a game world with floating blocks over a checkered baseplate) in saturated colours with thick outlines, nothing photographic.
Motion: something moves on every line, and the picture changes every 1-3 s.
- Things arrive with a squash (0.72, 1.12, 0.95, 1 over 0.3 s), an overshoot ease or a white poof cloud, never a plain fade.
- The camera pushes in 5-15% per shot and shakes (decaying over 0.35 s) on every hit.
Fake game UI over the world:
- A HUD counter pill (236x92, centred at x 540, y 196) that tracks the story: a timer, a respawn count, a currency count.
- "[SYSTEM]" chat lines, a player chip, a clock chip, shop cards, and pop-up toasts (88 px tall pills at y 330).
- A SUBSCRIBE button with a cursor hand.
Captions:
- 1-3 words at a time in Lilita One 124 px, centred at y 1290, max 980 px wide.
- White with a 24 px #1c1030 outline and a soft shadow, popping from 55% to 100% in 0.16 s.
- Key words in caps and colour: red #ff5165, green #5dff8f, yellow #ffd23f, pink #ff7ad9, cyan #7debff, orange #ffa53d, purple #c49bff.
- No full stops on screen, ever.
Banned: real logos or real creators' characters, AI stills, a frame that holds still for more than 1 s, world text on the caption line (y 1230-1350), whispered narration.
</direction>

<structure>
Length: 40-45 s, 85-110 words. Times below are from a 41.8 s Short; scale them to the voice.
0-2.3 s: the narrator reads the title as the first line ("What if your mom owned Roblox?") over the most striking picture of the rule (a chair spins round to reveal Mom, a bed turns into a spawn point). Caption words punch in.
2.3-4.5 s: one character line that proves the rule ("Hi sweetie! I bought it!"), with a toast pill.
4.5-10 s: DAY 1. A card pops: 230 px Lilita One in yellow #ffd23f with a 38 px outline, tilted -6 degrees. It comes in over 0.22 s and leaves at 1.05 s. One picture, one small joke, a reaction line ("Noooo!").
10-15.7 s: DAY 2, bigger. The HUD counter appears and ticks with a green "+1".
15.7-20.5 s: DAY 3, the biggest, with a callback to an earlier character.
20.5-23.5 s: the subscribe beat told inside the story ("Subscribe, and Mom adds one more minute"). The button is pressed at about 21.3 s.
23.5-33 s: the turn. The kid fights back and wins: a fast 3-shot montage (0.85 s each on bursting colour backgrounds), the counter racing to 1,000,000, confetti.
33-40.3 s: the payoff. The other side wins after all, in one clear picture and one line.
40.3-41.25 s: a short tag from the kid ("Wait, what?").
41.25-41.8 s: REWIND. The whole film scrubs backwards to frame 0 in 0.55 s with scanlines and a REWIND label, so the last line leads straight back into the first. No "comment X" ask.
</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 and lines: "you" and "your", the title as line 1, one picture per line, 10-16 words at most per line, words a 10-year-old uses, no names.
   - Make Roblox itself the topic in every beat (Robux, servers, obbies, avatars, the shop), not just a word in the title.
   - Days 1-3 escalate. The turn and the payoff flip it twice.
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: an energetic young man (median pitch 160-210 Hz). Kid: a high cartoon voice (about 340 Hz). Mom: bright and sassy (about 280 Hz).
   - Use audio tags: [excited], [screaming], [sweetly], [mischievously]. Cost: about 750 credits, about $0.14 for a 42 s Short.
   - Free route: Kokoro or Piper locally. Pick the liveliest voices; it will sound flatter.
   - Cut the takes into lines with faster-whisper word times (pad 0.06 s before and 0.15 s after). Squeeze pauses inside a line to 0.22 s. Leave 0.2 s between lines and 0.3-0.35 s before each DAY line.
   - 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 1.07; past that it sounds sped up.
   - Run whisper again on the finished track for word times. Every cut, card, 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: window .seek(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, squash, rotation, arm and leg angles, head tilt, eyes, brows, mouth, look.
   - Find hand positions with the same arm maths, so held props (a tablet, a sock, a money bag) sit in the hand.
   - Lilita One (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, characters cut off, cards over faces and empty corners.
6. Render (free): seek every frame in headless Edge or Chromium and encode inside the page with WebCodecs H.264 on the GPU: 60 fps, 60 Mbps, 2 motion-blur subframes over half a frame.
   - It renders at about 18 fps, so 42 s takes about 2.5 min.
   - Re-encode with libx264 CRF 20, preset slow, for the upload (about 27 MB).
7. Sound (free):
   - Music: one playful bed (Mixkit, free licence). Cut it on the two big moments (the shutdown and the twist).
   - Effects: Mixkit, placed so each effect's loudest moment lands on its frame. Use a whoosh and a pop on cards, a poof, a coin ding on +1, a boing on jumps, a slam, a glint, a party horn, a sting, a rewind.
   - 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.
   - What worked: the music bed about 19 dB down, effects that land on speech 12 dB down, effects in gaps 4 dB down, and every cue capped at -3 dBFS peak. Result: minimum 15.5 dB, median 17.9 dB.
   - Two-pass loudnorm to -14 LUFS, true peak -1.5.
8. Cover (free):
   - Study the top 5-10 rival covers first. On YouTube the custom cover is the middle 405x720 of the video's maxresdefault.jpg.
   - Background: a bright radial sunburst (#f6a8ff to #a24dff to #3d138f, with 22 white rays at 13%).
   - Title, 2 lines: line 1 at 300 px in a pink gradient (#ffe3f4 to #ff4fae). Line 2 at 140 px in a yellow gradient (#fff27a to #ffa51a), ending in "?!". Each letter bounces and is tilted -4 degrees, with a thick dark outline.
   - The main character big and smug with one prop circled in red (15 px #ff2d3d), the kid shocked in front, both cropped by the frame edges. Draw it with the same rig.
   - Check it side by side with the rivals at 240 px wide. JPG under 2 MB.
9. Upload text:
   - Title: "What If Your MOM [did X] Roblox?" plus 1 emoji, 28-49 characters, one word in caps.
   - Description: 2-3 teaser lines with caps words and emoji, ending on a question, then 5 hashtags and a line saying the video is not affiliated with Roblox.
   - Decide YouTube's made-for-kids setting yourself before posting: it changes comments and reach.
</build>

<gotchas>
- Plain script tags share one global scope, so a top-level const declared in two files is a SyntaxError that blanks the whole page. Give each file's names their own prefix.
- Whisper writes "8" for "eight" and "No!" for "Noooo!". Match lines by word count, not exact text.
- Whisper's word times drift up to 0.3 s where one character hands over to another. Time those lines from your own assembly list.
- On the first mix the narrator was only 6 dB over the music: the voice had hit the -3 dBFS peak cap and could not go louder. Pull everything else down instead, and measure before you normalise.
- A raised arm drawn behind the head hides the hand holding a prop. Pick arm angles where the hand clears the head.
- A camera that follows a jumping avatar bounces with it. Follow x, and ease y on its own.
- World text on the caption line (a nameplate, a sign) collides with the captions. Keep it above y 1230.
- A DAY card covers a head in game shots. Frame heads below y 600 while a card shows.
- Reading pixels back every frame (toDataURL) knocks the canvas off the GPU and renders about 50x slower. Encode inside the page instead.
- The 60 Mbps master is about 300 MB for 42 s. Render it to a big drive and upload the CRF 20 copy.
- The first version ran 55 s and ended on "Comment BED". The biggest rivals run 43-46 s, and a comment ask is wasted if comments end up off. Cut to 40-45 s and end on a loop.
- Stock narrators read at about 2.2 words a second; the rivals talk at about 3.1 while talking. Ask for an energetic read and squeeze pauses before you reach for atempo.
</gotchas>

<start>
Ask me for my story idea, or pitch 3. Then show me the script (word count and seconds per line) and the shot list before any voice is generated.
</start>

What you need

  • Coding agent with a shell; Python 3, Playwright with Chromium or Edge, FFmpeg, faster-whisper, numpy and Pillow (the prompt installs what's missing).
  • A voice: Kokoro or Piper locally for free, or ElevenLabs (about US$0.15 a Short).
FILES

Download the result

Technical details and provenance
Provider
Anthropic
Model
Claude Opus (creator-reported)
Published
Publication
community