What is motionmaxxing?
motionmaxxing, by Tejas Makwana (Tejashmakwana), is an Agent Skill that has a coding agent direct and build short motion-graphics films: launch videos, product promos, brand stings and kinetic type. It was published in a single upload: the repo was created on 6 October 2026 and its two commits both landed that evening, with nothing pushed since and no GitHub releases. The README itself calls it an early release.
The film is a plain HTML page driven by motion.js, a seek-safe runtime on top of GSAP with 12 named eases fitted to 4,288 moves measured frame by frame from 53 professional launch moments. The skill's own render.mjs seeks each frame in headless Chrome and encodes with ffmpeg; the description says it works in HyperFrames or plain HTML, and it is not a Remotion or After Effects tool. Given a URL it captures the brand's colours, fonts, logo and screens, and scripts then score the render against six gates, G0 to G5. Voice, sound effects and music are optional through an ElevenLabs key.
The repo is Apache-2.0 with about 1,200 GitHub stars and 113 forks. It installs by cloning into ~/.claude/skills and needs Node 22 or later, Python 3, ffmpeg and Google Chrome. The bundled GSAP 3.12.5 keeps its own GreenSock Standard License, which the README notes is not open source. The skill says it lives inside FinalMotion, the author's product.
Is motionmaxxing worth it?
motionmaxxing is the rare video skill whose rules come from measurement rather than taste. The author recreated 53 three-to-seven-second moments from designer-made launch films as frame-accurate GSAP timelines, seeked every frame and collapsed the result into 4,288 distinct moves. Out of that came the numbers the skill enforces: a median move of 10 frames (0.33 s), landings that ease out 58% of the time, overshoot on under 9% of moves and late when it happens, logos that crash in from about 3.5 times their size, and 3,392 transform entries against 584 opacity fades. Clustering the curves gave 12 eases, from softLand to accelExit, which the runtime ships by name and which, by the author's account, fit 89% of measured moves within an RMSE of 0.12.
The second half is a ban list, and it reads like hallmark's for video. SKILL.md lists 12 things a film may never ship: corner labels and counters such as "02 / SPEAK", a held left headline beside a device, a rounded UI card with a status dot, a headline typed with a caret, flat swatch grounds per scene, an end card held past 1.4 s and a film nobody looked at. A ten-step process asks for a storyboard and three kinds of idea before any build. Then render.mjs produces the film, look.py computes gates G0, G2 and G3 from the frames, lint.mjs fails page chrome in the DOM as G5, and a mandatory NOTE.md must quote those numbers instead of declaring the film good. G1 and G4 have no script and are judged by eye, which the skill makes the agent admit.
The cost is weight. SKILL.md alone is about 20.6 KB, and step zero adds the two taste files and a skim of judgment.md, so roughly 53 KB or 13,000 tokens at four characters to a token load before the brand is captured. The 14 reference files total about 452 KB, over 110,000 tokens if an agent reads them all, and the author publishes no token or time cost for a film. Rendering is slow where it matters: hero3d scenes in WebGL run at about 7 to 11 fps at 1080p on an Apple GPU, according to KNOWN-LIMITS. Only Claude Code has documented install steps; other agents get a folder to point at.
The evidence is thinner than the polish suggests. The blind A/B record behind the taste layer has one human rater, the 9 of 10 picks for the skill came from an earlier generation, and no A/B exists for this one. The 716-row precedent library was decoded by model analysts and is marked unverified. The measured moments are SaaS-style launch clips, so the statistics describe that genre. Most of all, nothing has been pushed since 6 October and a 16 October issue about Chrome processes left behind on macOS is still open. Against openmontage, which plans whole productions across a hundred tools, motionmaxxing does one thing deeper; against hyperframes, it is a director sitting on a near-identical Chrome and ffmpeg render path rather than a renderer. As a reference for how agent-made motion should move, it is worth reading even if it never gets a second commit.
Where it earns its place
A founder at a five-person SaaS start-up with a launch next week
Turn the product's website into a 20-second launch film without hiring a motion designer
Run /motionmaxxing in Claude Code with the site URL; brand.mjs captures colours, fonts, logo and screens, and the agent shows the idea and storyboard before building. Free apart from agent tokens and, for voice and music, ElevenLabs usage.
A frontend developer whose agent-made promo looks like a moving slide deck
Find out exactly what is wrong with a draft and fix it
Hand the skill draft.mp4 and its index.html and ask for a slop, mid or maxxed rating; look.py and lint.mjs print which of G0, G2, G3 and G5 fail, and the agent fixes in a fixed order. The lint cannot see text drawn in a canvas or a generated image.
A developer building their own video agent on HTML and GSAP
Borrow measured timing instead of guessing durations and curves
Load runtime/motion.js and call the 12 named eases such as M.ease.softLand with the measured defaults, and query library/moments.jsonl through precedent.py per beat. GSAP's own licence still applies to the bundled copy.
Before you commit
- Setup
- git clone the repo into ~/.claude/skills/motionmaxxing, then bash install.sh, which checks requirements and installs nothing system-wide. Needs Node 22+, Python 3, ffmpeg and Google Chrome; scripts/providers.sh reports what the machine has. Invoke with /motionmaxxing or a plain request.
- Integrations
- Claude Code is documented; any agent that reads a SKILL.md folder and has shell access can use it. Renders through its own Chrome and ffmpeg pipeline, can work alongside HyperFrames, vendors GSAP 3.12.5 and Three.js r186, and optionally calls ElevenLabs for voice, sound and music and the Codex CLI for generated surface plates.
- Data and privacy
- Runs locally. Brand capture fetches the target site through headless Chrome; scripts for voice and music send text to ElevenLabs only when ELEVENLABS_API_KEY is set, and image plates go through a Codex login only when asked. The README describes no telemetry. Apache-2.0, with GSAP under the GreenSock Standard License, Three.js and opentype.js under MIT and Inter under SIL OFL 1.1.
- Cost at scale
- No paid edition in the repo. The cost is agent tokens (about 13,000 tokens of instructions before work starts, more as references load), local render time, ElevenLabs usage and about 1 to 2 minutes per generated plate.
- Limits
- No commits since 6 October 2026 and one open issue. G1 and G4 are judged by eye; lint.mjs misses text in canvas, WebGL, images and video; WebGL renders at about 7 to 11 fps at 1080p; ElevenLabs music may miss tempo or length; fetch_logo.mjs can return a header ribbon instead of the mark.
What’s good
- Eases and timings measured from 4,288 professional moves
- Concrete never-ship list aimed at slide-deck video tells
- Scripted gates and a lint check the render; NOTE.md must quote them
- Seek-safe runtime with a self-test, so renders are reproducible
What’s not
- One upload on 6 October 2026, no commits or releases since
- Taste layer rests on a single rater; no A/B for this generation
- Heavy setup: Node 22+, Python 3, ffmpeg and Chrome
- Large instruction load and no published token cost
Best for
Developers who already make product films with an agent in HTML and GSAP and want the output to move like a motion designer's, and who can live with a repo that may not be maintained.
motionmaxxing alternatives
motionmaxxing: common questions
- Is motionmaxxing worth it?
- motionmaxxing writes down more real motion-design craft than any video skill we have read, then leaves it as one unmaintained upload checked by a single rater. We rate it 3.6 out of 5.
- How much does motionmaxxing cost?
- Free and open source (Apache-2.0; bundled GSAP under its own licence). An optional ElevenLabs key enables voice, sound effects and music, billed by ElevenLabs.
- What are the downsides of motionmaxxing?
- One upload on 6 October 2026, no commits or releases since; taste layer rests on a single rater; no A/B for this generation; heavy setup: Node 22+, Python 3, ffmpeg and Chrome; large instruction load and no published token cost.
- Who is motionmaxxing best for?
- Developers who already make product films with an agent in HTML and GSAP and want the output to move like a motion designer's, and who can live with a repo that may not be maintained.
- How much does motionmaxxing cost for a team or in production?
- No paid edition in the repo. The cost is agent tokens (about 13,000 tokens of instructions before work starts, more as references load), local render time, ElevenLabs usage and about 1 to 2 minutes per generated plate.
- Is motionmaxxing safe for company data?
- Runs locally. Brand capture fetches the target site through headless Chrome; scripts for voice and music send text to ElevenLabs only when ELEVENLABS_API_KEY is set, and image plates go through a Codex login only when asked. The README describes no telemetry. Apache-2.0, with GSAP under the GreenSock Standard License, Three.js and opentype.js under MIT and Inter under SIL OFL 1.1.
- What are the best motionmaxxing alternatives?
- The closest alternatives we have reviewed are HyperFrames, OpenMontage, Hallmark and ffmpeg-skill.
Answers checked Oct 31, 2026.