On October 8, 2026, Anthropic opened a beta of Claude Motion, a tool that turns a report, a chart or an image into a short explainer video. It does not use a video generation model. Claude writes code that animates text, charts, shapes and images, and that code plays back like a video. It is the choice I would have made. A video described by parameters can be fixed one value at a time, while a video from a generation model has to be generated again from scratch. The trade-off decides how to use it well: that flexibility lives in the artifact, and the exported MP4 does not keep it.
What Anthropic announced
The facts below come from two sources: the launch article Anthropic posted on claude.com on October 8, 2026, "Build live dashboards and animate explainers with Claude", and the help page "Get started with Claude Motion". Motion shipped the same day as Claude Dashboards, which builds dashboards wired to live data. Anthropic's examples show the intended ground: a 30-second explainer for an all-hands, an animated chart in a board deck, a product walkthrough for customer onboarding. Motion is in beta on the Team and Enterprise plans. It is on by default for Team, and off for Enterprise until an owner turns it on under Organization settings, in the Artifacts section. Unlike Dashboards, which is also open to Pro and Max, it stays limited to team plans. There is no separate price: it draws on the plan's usage limits, and a long or complex video uses more of them than a regular message.
How you use it
You start Motion from any conversation, either by asking for a video or by picking Motion from the Output menu in the message box. You can also start from a template in the Artifacts tab. The help page recommends saying who will watch, where it will play and roughly how long it should run, and attaching the source material: the report, the spreadsheet, the visuals. Claude builds the video and saves it as an artifact. You then revise it by asking in the conversation, or by hand in the editor. When it is ready, it exports as an MP4 or opens in Adobe, Descript, HeyGen, Higgsfield, invideo, Luma AI or Runway, with Canva and Captions announced for later. A video is private by default and is shared through a link.
What Anthropic says about how it works, and what they leave out
Anthropic states one thing and repeats it in every document: Claude writes code that animates text, charts, shapes and images. They do not say which code. The language, the animation library, the frame rate and resolution of the MP4, the maximum length: none of it is published, and I am not going to guess. The principle itself is familiar to developers. Remotion, an open-source framework, describes a video as React components and renders it frame by frame. Nothing says Motion is built on it, but the idea is the same: the video is the output of a program, and the program is what you edit.
Why a video model thinks differently
A video generation model produces pixels. You describe a scene and it synthesizes a sequence of frames that matches it. In the result, an 18% figure does not exist as a number. It is a patch of color shaped like one. Changing it to 19% means running the generation again, and the new take also differs elsewhere: a framing, a color, a motion. Motion produces a structure. The figure is a value, with a position, a start time, a duration and an easing curve. The help page puts it this way: you change one detail, a word, a number or a timing, and the rest stays as it was. For a quarterly report that is the property that matters, because the numbers often move until the night before the meeting. Anthropic is also upfront about what the tool does not do: no realistic footage, no generated people. The export list includes Runway, Luma AI and Higgsfield, three video generation tools, which shows Motion is meant to sit upstream of them and feed them.
The same split, on this portfolio
The home page of this site uses both approaches, and they are maintained in opposite ways. The arrival curtain, "ENZO RUFFIN" with an image block opening between the two halves of the name and a run of hard cuts, is a GSAP timeline. Its pacing lives in an object of constants at the top of the file: 0.9 seconds for the letters to rise, 0.42 seconds between two cuts, 1.55 seconds for the final expansion. Making the cuts faster means changing one number and reloading the page. The compass in the Networks section is a sequence of 144 WebP frames at 720 by 1280 pixels, 2.5 MB in total, painted on a canvas as you scroll. Changing the drawing of the needle would mean redoing all 144 frames. And to make those pixels play without stutter, I had to build a whole mechanism around them: decoding upcoming frames ahead of time, smoothing computed by hand in the ticker, snapping to a whole frame at rest so two ghost strokes are not left overlapping. The timeline needs none of that. It is the gap between Motion and a video model, brought down to the scale of a single page.
The MP4 is a build, the artifact is the source
The code only protects you while you stay inside Motion. The exported MP4 is pixels again, with the same constraints as a generated video: a wrong number in the file gets fixed by going back to the artifact and exporting again. So I would treat the export as a build. The artifact is the source, kept and named like a project file, and every revision goes through it. Editing in another tool comes last, once the content is locked. The second rule is more down to earth: an editable number is not necessarily a correct one. Claude reads the report you give it and stages it. Every value on screen gets checked against the source, the way you would check a slide before putting it in front of leadership.
What is left to measure
Motion is in beta, limited to team plans, and Anthropic has published no technical detail about the render. The question I am keeping open is the finish: whether the easing curves and the pacing between shots hold up against a timeline tuned by hand, like the arrival curtain, where every cut was set to the hundredth of a second.