Opus 5.5 Motion Design: How Claude Writes Motion Graphics

Anthropic's launch post does not mention motion design at all. It pitches Opus 5.5 on agentic coding, cost and clearer writing, and the closest it gets to visuals is a tester who had several Claude models build a game from one prompt, where Opus 5.5 "scored higher than any other model on the strength of its graphics and polish."

The motion design part came from users. Within days, X was full of launch films, showreels and explainers that creators said were made by prompting the model. Revid, a video tool company, collected the ones that passed 100 likes during launch week into a public gallery of 64 videos and credits each to its creator. Its top three alone drew 6.9 million views.

So when people search for claude motion design or claude motion graphics, they usually want one of three things: to see the examples, to copy the prompt and setup, or to find a way to get the same look without coding.

Every serious workflow in the gallery follows the same loop. You describe the video, the model writes a scene as code with every element timed to the frame, a local preview plays it, you ask for changes, and finally a renderer captures each frame and encodes the file. Because text and shapes are drawn by a browser engine rather than generated as pixels, a 12-point caption stays as sharp as the headline.

Two open toolchains came up again and again:

Remotion's own guide is titled "Prompting videos with coding agents" and names Claude Code first among supported agents. HyperFrames, from HeyGen, describes itself in one line: "Write HTML. Render video. Built for agents." Several gallery entries are labelled Claude Code plus Opus 5.5 or HyperFrames, which matches what the creators themselves said they used. When people search for a claude motion graphics skill, these skill packs are usually what they mean: instructions the agent loads so it writes valid code for that renderer on the first try.

Revid published a case study that measured every cut, colour and sound in the 63 launch-week videos that passed 100 likes on X. Its findings, as reported by Revid, are worth knowing before you design anything:

• The typical hit ran 29 seconds, had 3 hard cuts, used music with no voiceover and showed readable text within two seconds.
• Showreels won the week: 8 of the 10 showreels ran exactly 15.1 seconds, 9 sat on a dark background, and the 10 showreels took 42% of all likes.
• Palettes were mostly black, white and grey with one accent colour.
• Craft did not predict likes. Posts that named Claude got 3.5 times the median likes, which Revid attributes to reach, not editing.

Revid is clear about the limits: every video in the sample was already a hit, so the study compares hits with bigger hits. Use it as a style brief, not a formula.

This is the Remotion route, taken from Remotion's documentation. The HyperFrames route is similar: install the plugin in Claude Code and start with its quickstart at hyperframes.heygen.com.

1. 1

   Get the prerequisites

   A coding agent such as Claude Code and Node.js. Remotion notes that most coding agents require a paid subscription.

2. 2

   Create a blank project

   Run npx create-video --yes --blank my-video, then move into the folder.

3. 3

   Add the agent skills

   Run npx remotion skills add so the agent knows Remotion's APIs, then npm run dev to open the preview.

4. 4

   Start the agent and prompt

   In a second terminal, start claude in the same folder and describe the video. Ask for a storyboard first and approve it before any animation code is written.

5. 5

   Iterate in the preview, then render

   Request changes one at a time (timing, colour, one scene) and render the final MP4 when it looks right.

The gallery prompts that travelled best share a structure: format and length first, then palette and type, then the beats in order, then what to avoid. Here is an original example in that shape:

Make a 15-second 16:9 motion graphics showreel for a fictional note-taking app called Ledger. Near-black background, white and grey type, one coral accent. Beats: 0-3s the word Ledger builds letter by letter; 3-8s three feature cards slide in with one-line captions; 8-12s a bar chart grows to show notes per day; 12-15s logo lockup and the line Write less, remember more. Hard cuts only, no voiceover, leave room for music. Show me the storyboard as a list before you write any code.

Ask for the storyboard first because changing a list is cheaper than changing code. Then keep each follow-up to one change, so you can tell which instruction caused which result.

On the API, Anthropic lists Opus 5.5 at $4 per million input tokens and $20 per million output tokens, with a fast mode at $8 and $40 (model docs). In the Claude apps it is available on the Pro, Max, Team and Enterprise plans. How much a single video costs depends on how many rounds of edits you run and how long the code gets, so nobody can quote a fixed per-video price honestly. Rendering itself happens on your machine with Remotion or HyperFrames.

If the terminal is the part you want to skip, the same idea, motion graphics as structured code rather than generated pixels, is how LazyKiwi Motion Studio works. You describe the clip in a sentence, approve a storyboard with scenes, timings and a palette, and get back an editable document in which every headline, colour and beat is a labelled control. It renders in the browser, so tweaks do not cost a new generation.

• For a single animated line or title card, the animated text generator has 11 presets and exports at four ratios.
• For a logo sting or channel intro, logo animation places your mark in 18 scenes.
• To start from a finished layout, browse the motion graphics templates and edit the copy and colours in place.

Free exports are 720p with a small LazyKiwi mark; any paid plan removes it and adds 1080p, per the Motion Studio page. The trade-off against the code route is control: a coding agent can build any scene you can describe, while a studio gives you a faster path inside its templates and engine.

They solve different problems. Code-based motion is the right tool when the content is type, logos, UI, charts or anything that must be exact and editable. A video model is the right tool when you need realistic footage: people, places, weather, camera moves through a real-looking scene. Video models such as Veo 3.1 Lite generate those pixels from a prompt or an image, but they are not built to keep a 12-point legal line readable or a logo's exact colour. Many of the gallery's launch films combined both: generated footage in the background, coded type on top.

Not directly as pixels. Claude writes the animation as code and a renderer produces the video file, which is how most of the gallery was made; Revid's own FAQ describes Opus 5.5 as writing "the animation as code". Any capable coding assistant, ChatGPT included, can attempt the same Remotion or HyperFrames workflow, since both tools document support for several agents. The launch-week examples were attributed to Opus 5.5, which is why the trend carries its name.