huashu-motion is a real MIT repo at github.com/alchaincyf/huashu-motion, but its "no video AI generator" pitch describes Claude Motion too: Anthropic's launch page says Motion "doesn't use a video generation model"

The artifact is a 48.088005 s MP4 (QuickTime/MOV container, mov,mp4,m4a,3gp,3g2,mj2), 1080x1920 portrait, HEVC Main profile at level 150 in yuv420p, 60 fps, 2,885 video frames, video bitrate 607,497 bps, audio AAC HE-AACv2 at 44,100 Hz stereo and 48,162 bps, overall 664,624 bps, 3,995,059 bytes on disk, yt-dlp format id bytevc1_1080p_664610-1; the video stream runs 48.083333 s and the container 48.088005 s. It was posted at epoch 1792646969, which is 2026-10-22T05:29:29Z, by handle @chase_ai_ (uploader id 7334112384573096990, nickname "Chase AI"), credited to "original sound" by Chase AI, with a 39-character description identical to the title, "an open source version of Claude Motion". At the capture time recorded by the metadata file's mtime (2026-10-22 18:04:03 local) the counts stood at 13,400 views, 818 likes, 164 reposts, 987 saves and 0 comments. I read 15 frames from a fresh 2 s extraction in tokdoc_frames/ (001, 002, 005, 006, 008, 009, 011, 014, 017, 018, 019, 020, 021, 023, 024) against a 157-word transcript in 9 segments. Every frame is the same split screen: the creator talking to camera on the lower half, and on the upper half a GitHub file view with the tabs "Preview | Code | Blame" and the buttons "Raw", copy, download and edit, scrolling through a README whose H1 reads "huashu-motion" and whose pull quote reads "Commercial-grade product launch films, in one go." Under it sit three badges, "License MIT", "Agent Skills Standard" and "Runtime Claude Code · Codex · WorkBuddy · Grok Bot · Muse · Dots". Frame 001 shows a rebuilt terminal typing /huashu-motion 给我的产品做一支发布片; frame 009 shows README prose reading "the full Grok Bot, huashu-design and DeepSeek films are on the same page" above a section headed "Point at the frame" with a code diff containing roomCam.set({ focus: [1050, 712], zoom: 2.8 });; frame 011 shows the paragraph "Press C, click the frame, type the change" and a timeline strip labelled hook, night, work, turn, burst, stop, dawn, mark, end; frame 014 shows the disclaimer "(The editor and terminal in this clip are rebuilt from the real interface; code details are illustrative. The comment is a real one.)" and the heading "Score first, then the picture"; frame 017 shows a yellow MORPH card captioned "20 种动画语言" with the line "Sixteen shots from seven films made the same day, cut on the beat."; frame 018 shows a card reading "23 种音效，踩在每一刀上". The spoken argument is that Claude Motion is locked behind Team and Enterprise, that a GitHub repo called Huashu Motion is an open source equivalent, that it draws motion in code rather than calling a paid video generator, and that the creator ran it against Hyperframes and preferred it for product demos. No terminal session, no render, no editor recording and no side-by-side comparison with any other tool appears in any frame I read; the only thing on screen for all 48 seconds is the project's own README. On the transcript-end check, the final segment ends at 47.90 s against 48.088005 s of audio, so nothing is timestamped past the end of the file and no part of this transcript is hallucinated.

The repo name survived Whisper intact, and the burned-in caption is the thing that breaks

Whisper usually mangles product names in this corpus, so "Huashu Motion" looked like a corruption. It is not. At 9.04 s the audio says "GitHub repo. It's called Huashu Motion", and the README H1 on screen in frames 001, 002, 005, 006, 008, 019, 020, 021, 023 and 024 reads huashu-motion in plain text. The GitHub API returns the repository as alchaincyf/huashu-motion, created 2026-10-18T12:19:27Z, MIT licensed, primary language JavaScript, with topics agent-skills, animation, claude-code, codex, gsap, launch-video, motion-graphics and skills. At the time I queried the API it carried 1,435 stars and 139 forks. The skill manifest at skills/huashu-motion/SKILL.md names the author as 花叔 and the homepage as https://github.com/alchaincyf, which is where the romanisation "Huashu" comes from.

The failure is in the other direction. At 44.04 s the creator says "Just Google Huashu Motion GitHub, and it will be the first thing that pops up", and TikTok's burned-in auto-caption renders that on screen in frame 023 (about 44 s) as "just Google Huawei". A viewer who reads the caption instead of listening searches for a Chinese telecom manufacturer. The same caption track mangles the audio's "SeedDance" into "seedance" in frame 014 (about 26 s), which happens to be closer to the correct spelling than the spoken word. On the underlying claim, the search I ran for "huashu-motion github" did return github.com/alchaincyf/huashu-motion as the first result, so the instruction works if you spell it the way the audio says rather than the way the caption shows.

"The Teams or Enterprise plan, which is the only place you can get it" holds up, with one word corrected

At 4.78 s the video says Claude Motion requires "the Teams or Enterprise plan, which is the only place you can get it". Anthropic's own launch article, published 8 October 2026, states: "Claude Motion is in beta on Team and Enterprise." The plan is named Team, not Teams, and the restriction is accurate as written on that page. Dashboards, announced in the same article, sits on a different line: "Claude Dashboards connects to a data platform like BigQuery, Databricks or Snowflake, or a CRM tool like Salesforce, and is in beta on paid plans."

The opening phrase needs a separate clarification. At 0.00 s the video says "the new Claude Motion design", which runs two product names together. Anthropic shipped Motion into beta and took three other surfaces out of beta on the same day. The article says "Docs, Slides, and Design are out of beta and available on every plan, including Free", and that "Since September 16, Design also works inside every Claude conversation, where it can use your context, files, and connectors." Claude Design is therefore available on the free tier, and the Team-and-Enterprise restriction applies only to Motion. If a viewer hears the first sentence as "Claude Motion Design is Team and Enterprise only", half of that is wrong.

"You aren't paying for any video AI generator" is true of huashu-motion, and equally true of Claude Motion

Between 14.80 s and 33.88 s the video builds its central contrast: "All the motion graphics are created with code, specifically things like JavaScript. So you aren't paying for like any video AI generator. Like we're not using Kling, we're not using SeedDance. This is all code."

The first half checks out against the repo. The English README's "How it works" section states: "A film is one index.html. GSAP, Canvas, Three.js, SVG, anything a browser draws. No video model; every frame is a function of the code and the time." The same section describes the renderer as "Several Chrome instances at once (by CPU count, up to 8 by default), picture and audio encoded in parallel", and the requirements list asks for Node 22 or later, a Chromium-family browser and FFmpeg, noting "no renderer is downloaded". The "Limits" section says outright: "It doesn't generate people or live footage. Everything is drawn in code; bring your own footage and it composes and packages it." The GitHub API reports the repository's primary language as JavaScript, and gsap is one of its declared topics.

The contrast against Claude Motion is where the video goes wrong. Anthropic's article says of Motion: "Claude writes code that animates your text, charts, shapes, and images, so you can change any word, number, or timing", followed by "It doesn't use a video generation model, so there's no generated footage and no AI-generated people." Claude Motion and huashu-motion use the same approach. Running the comparison through Kling and Seedance sets up a difference that does not exist between the two products the video is actually comparing. The real gap between them is the plan requirement and the license, not the rendering method.

The two named generators are real, and both names need correction. Kling is a video generation model from the Kuaishou AI team, first released in June 2024, and the current line is Kling 4.0. Seedance is ByteDance's video generation model from its Seed team, with Seedance 2.0 as the current release. The audio's "SeedDance" with a doubled D is not the vendor's spelling; the product is written Seedance.

Transitions, music and a browser editor: two are documented features and one is not

At 26.48 s the video lists three things the repo adds: "It adds transitions, it adds music, and it even includes an editor that you can use inside of your browser." Each deserves a separate verdict against the README.

Music is documented in detail and is the strongest of the three. The README says: "For launch films the agent writes a score in ABC notation, with marks that land on the scenes, and a sampled grand piano, string sections and drum kits play it (after a one-time om samples build; otherwise synths). Each landing gets a sound effect on its frame. All of it is generated locally; no paid API." The feature list counts "12 product types × 3 fallback music directions, 23 sound effects placed on the line where the motion lands". Frame 018 confirms this on screen with the card "23 种音效，踩在每一刀上", which reads as twenty-three sound effects landing on every cut.

The browser editor is documented. The command table lists om studio as "Live preview, scene strip, direct edits, frame comments", and the README describes the interaction as pressing C, clicking a frame and writing a note, with version history and ⌘Z undo, and with "Titles, numbers, colours and hold times show up as fields you can edit yourself." Frame 011 shows that paragraph and frame 021 shows the studio's input bar. One caveat belongs here: the README's own disclaimer in frame 014 says the editor shown in the animated demo is "rebuilt from the real interface" and that "code details are illustrative", so the GIF is a recreation rather than a screen recording.

Transitions are the weakest of the three. The word "transitions" appears exactly once in the English README, and it appears inside a quoted complaint from the author's own revision log: "And the workflow part is clear, but the 1, 2, 3, 4, 5 transitions feel very much like PowerPoint, not premium enough!" The repo does document something adjacent, "20 motion languages" derived from measuring "21 real launch films (cuts, beats, loudness, energy curves) and 142 high-engagement launch and motion films", plus cut timing driven by the score, which frame 017 shows as "20 种动画语言" and "Sixteen shots from seven films made the same day, cut on the beat." Calling that a transitions feature is a reasonable paraphrase of cutting on the beat, but the repo does not ship a named transitions library, and the only time the README uses the word it is describing a result the author rejected.

The Hyperframes head-to-head is the creator's private benchmark, and nothing about it appears on screen

At 33.88 s the video says "I also ran it head to head with Hyperframes, and this Huashu Motion did a little bit better when we're looking at things like product demos."

Hyperframes exists and the name is spelled HyperFrames by its maker. The repository is heygen-com/hyperframes, created 2026-03-10T01:51:06Z, Apache-2.0 licensed, written in TypeScript, with the description "Write HTML. Render video. Built for agents." and the homepage hyperframes.dev. At the time I queried the API it carried 69,979 stars. The product page at hyperframes.heygen.com describes it as "The video editor built for agents" and attributes it to HeyGen. It renders video from HTML rather than from a generative model, which puts it in the same category as both huashu-motion and Claude Motion.

The comparison claim is Unverified. The creator names no rubric, no prompt, no number of trials, no scoring method and no output. I read frames 017, 018, 019, 020 and 021, which span roughly 32 s to 40 s and cover the entire head-to-head sentence. Every one of them shows the huashu-motion README and nothing else. Frame 019, at about 36 s, carries the burned-in caption "head with hyperframes" over a README section showing a single orange 定位 card from huashu-motion's own demo reel. No HyperFrames output, no split screen, no two renders of the same brief, and no artifact of any comparison appears anywhere in the video. A viewer is being asked to accept a ranking between a 1,435-star project and a 69,979-star project on the creator's word alone, with no evidence on screen.

The chronology runs the way the video implies, and the repo's README says so in its own words

The phrase "an open source version of Claude Motion" at 9.04 s implies the repo came second. It did, and the dated record is unambiguous in both directions.

Anthropic published the Claude Motion announcement on 8 October 2026. The huashu-motion repository was created on 2026-10-18T12:19:27Z, ten days later. Its first commit, a831070, is dated 2026-10-18T12:19:18Z with the message "huashu-motion 0.1.0", and release v0.1.0 was published at 2026-10-18T12:26:46Z. The project moved fast after that: v0.1.1 on 2026-10-19, v0.2.0 at 2026-10-22T09:08:19Z, v0.3.0 at 2026-10-22T12:17:47Z and v0.3.1 at 2026-10-22T12:19:37Z. That last release landed about seven hours after this TikTok was posted, which means the README the creator is scrolling through on screen is a version that has already been superseded twice on the same day.

The author states the derivation himself. The README's "Why it exists" section reads: "On 8 October 2026 Anthropic released Claude Motion: Claude animates reports, charts and product explainers in code, and you point at a part to edit it. At launch it was a beta for Team and Enterprise only. In April I had turned Claude Design's core ideas into a skill, huashu-design; after Motion came out I gave Claude Code one line: our skill should fully reproduce Claude Motion, and be much stronger." The video's framing matches the author's stated intent. Worth adding for context: HyperFrames, the thing the creator says huashu-motion beat, was created on 2026-03-10, five months before Anthropic announced Motion and seven months before this repo existed. The open source HTML-to-video idea was not new in October 2026, and huashu-motion is the newest entrant in that category rather than its origin.

Key Takeaways

Verified: The repository is real, open source and findable. It is github.com/alchaincyf/huashu-motion, MIT licensed, created 2026-10-18T12:19:27Z, JavaScript, 1,435 stars and 139 forks at the time I queried the GitHub API, authored by 花叔 under the account alchaincyf.
Verified: Whisper did not mangle the name. "Huashu Motion" matches the README H1 huashu-motion visible in nine of the fifteen frames I read, and the GitHub API confirms it.
Verified: The repo renders from code, not from a generative video model. The README states "A film is one index.html. GSAP, Canvas, Three.js, SVG, anything a browser draws. No video model; every frame is a function of the code and the time", and the Limits section adds "It doesn't generate people or live footage."
Correction: The video's central contrast does not hold. At 20.90 s it says you "aren't paying for like any video AI generator. Like we're not using Kling, we're not using SeedDance", framing code rendering as what separates the repo from Claude Motion. Anthropic's 8 October 2026 article says Claude Motion "doesn't use a video generation model, so there's no generated footage and no AI-generated people". Both tools draw frames from code. The difference between them is the plan requirement and the license.
Correction: The caption on screen misspells the search term. At 44.04 s the audio says to Google "Huashu Motion GitHub"; the burned-in caption in frame 023 renders it as "Huawei". Searching the caption finds the wrong company.
Partial correction: The plan restriction is right, the plan name is wrong. At 4.78 s the video says "the Teams or Enterprise plan". Anthropic's wording is "Claude Motion is in beta on Team and Enterprise."
Clarification: The opening phrase "the new Claude Motion design" at 0.00 s merges two products. The same Anthropic article says "Docs, Slides, and Design are out of beta and available on every plan, including Free." Only Motion is restricted to Team and Enterprise.
Partial correction: Of the three features listed at 26.48 s, music and the browser editor are documented and transitions are not. om studio is listed as "Live preview, scene strip, direct edits, frame comments", and the score system is described in full. The word "transitions" appears once in the English README, inside a quoted complaint that some transitions "feel very much like PowerPoint, not premium enough!"
Correction: Seedance is spelled with one D. The audio says "SeedDance" at 26.48 s. The product is Seedance, from ByteDance's Seed team. Kling, the other named generator, is from the Kuaishou AI team and was first released in June 2024.
Unverified: The head-to-head against Hyperframes. The claim at 33.88 s has no rubric, no prompt, no trial count and no artifact. Frames 017 through 021, which cover the entire sentence, show only huashu-motion's own README. No comparison output appears anywhere in the video.
Context: HyperFrames is heygen-com/hyperframes, an Apache-2.0 TypeScript project by HeyGen, created 2026-03-10 with 69,979 stars at the time I checked. It is seven months older and roughly forty-nine times more starred than the repo the video says beat it, and it also renders video from HTML rather than from a generative model.
Context: The causal order in the video is correct and the author confirms it. The README's "Why it exists" section says the project started after Claude Motion's 8 October launch with the instruction "our skill should fully reproduce Claude Motion, and be much stronger."
Context: The README on screen was already out of date when the video published. The video posted at 2026-10-22T05:29:29Z; releases v0.2.0, v0.3.0 and v0.3.1 all shipped later the same day, the last at 2026-10-22T12:19:37Z.

Resources

alchaincyf/huashu-motion, GitHub establishes the repository's real name, owner, MIT license, 2026-10-18 creation date, star and fork counts, and the commit and release history I cite for the chronology.
huashu-motion README (English), GitHub establishes the "No video model; every frame is a function of the code and the time" wording, the om studio editor entry, the ABC-notation score system, the 23 sound effects, the 20 motion languages, the Node 22 and FFmpeg requirements, and the "Why it exists" account of the project starting after Claude Motion.
huashu-motion SKILL.md, GitHub establishes the author name 花叔, version 0.3.1, and the om command surface including studio, sound, render and check.
Build live dashboards and animate explainers with Claude, Anthropic, 8 October 2026 establishes the Claude Motion launch date, the sentence "Claude Motion is in beta on Team and Enterprise", the sentence "It doesn't use a video generation model, so there's no generated footage and no AI-generated people", and the line placing Docs, Slides and Design on every plan including Free.
heygen-com/hyperframes, GitHub establishes that HyperFrames is a HeyGen project, Apache-2.0 licensed, created 2026-03-10, written in TypeScript, with the tagline "Write HTML. Render video. Built for agents." and the star count I cite.
HyperFrames, HeyGen establishes the product positioning "The video editor built for agents" and HeyGen's ownership of it.
Seedance 2.0 official launch, ByteDance Seed establishes the correct spelling of Seedance and that it is a ByteDance Seed video generation model.