Motion graphics with Claude: code video or After Effects?

Since Opus 5.5 came out, X is full of motion graphics nobody keyframed by hand. Product films, animated Figma designs, whole short stories. Then people try the same prompt and get a centered headline on a gradient.

Lukas Margerie’s video How to make insane motion graphics with Opus 5.5 names the reason well: the prompt is about 10 % of the result. The other 90 % is the harness around it. Rules, brand files, references, a sound plan and a review loop.

We agree, and MotionUI is a harness too. The difference is where the film ends up. In the code approach it ends up as an MP4. With MotionUI it ends up as layers and keyframes in After Effects that you can still open. That one difference decides how the second round of client feedback goes.

How the code approach works

Opus 5.5 can’t output a video file. It writes code. Without further instructions it builds the whole film as HTML and SVG, often in a single file, renders it frame by frame in a headless Chrome and stitches the frames together with FFmpeg. With a framework like Remotion or HyperFrames connected, it writes for that instead.

The good results in the video all come from the same layers of instructions:

1. Studio rules that ban the generic look and define how rendering and review work.

2. Brand and spec files with logo, colors, fonts and a list of UI components like buttons, loaders and charts.

3. A reference film, cut into a frame every half second and turned into a style guide. The rule: take its grammar, never its content.

4. A beat grid, a table of every state on the music grid, approved before any code is written.

5. Sound in code: a synthesized 120 BPM track, a beats.json with every beat and downbeat, UI sounds generated in Node, and the animation retimed to the beats.

6. A critique loop where the model scores its own film per category and fixes everything below 7 or 8 out of 10.

The effort is real. One of the strongest films shown took 163 model calls and almost 7 hours. “One prompt” mostly means one long director’s brief.

Where it gets hard: the second round

The first version is rarely the problem. Client work lives in revisions. On a recent explainer film of ours, round one came back with 17 points: a gray that was too light, an icon that made no sense, a card that left a beat too early, a logo that sat 15 % too large.

In a code video, every one of those points is a new prompt and a new render. The model rewrites the file, and then you check whether the parts the client already approved still look the same. Usually they look close. Close is a problem when the client signed off on the rest.

The MP4 is the end of the line. There are no layers to hand to a motion designer who wants to fix one curve by hand. Fixes go back through the model, every time.

The model grades its own work. In the critique loop, the model decides what an 8 out of 10 looks like. In our projects, the model’s own “improvements” were often exactly what got rejected. On one sound mix, the version that won was a stripped-down copy of the first draft, after two “improved” versions were thrown out.

Synthesized music sounds synthesized. That works for a showreel on X. For a brand film with a voice-over, most clients expect a real track and sound effects that fit it.

The MotionUI approach: same harness, different output

With MotionUI, Claude Code doesn’t write a film from scratch. It drives the MotionUI plugin inside After Effects and builds with tested building blocks: presets, easing curves, sequencing, text effects, rectangle rigs, cursors, proximity effectors and matchcut transitions. The harness ideas from the video all have a place here, they just produce layers instead of code.

Your design is the source. The Figma plugin brings whole frames into After Effects and keeps the layout and layer structure. Text arrives as editable text layers with its fonts and sizes. The model animates the design you approved instead of rebuilding the UI from screenshots.

References become measured timing. Point it at a reference video and it measures shots, cuts, the length of each move and each hold. That goes into a saved style that describes the craft, never the reference’s logos or content. Same rule as in the video, stored so the next film can use it.

Review runs on rules from real revisions. Before anything is shown, a review step checks the comp for mistakes clients have actually flagged: elements visible before their card appears, off-center icons, uneven padding, abrupt or linear big moves, motion blur on still text. It returns a contact sheet and the measured motion, so the model fixes against facts, not its own taste.

The second round is an edit. Every layer and keyframe is real After Effects, and every change the model makes is a named undo step. The gray is one color value. The early card is one keyframe. Everything the client approved stays exactly as it was, and a motion designer can take over by hand at any point.

The final render is After Effects. Render queue, real motion blur, ProRes, and whatever your delivery pipeline already expects.

Side by side

The code approach wins on setup and scale. Nothing has to be installed, and a server can render ten variants overnight. MotionUI wins as soon as someone gives feedback.

What we took from it: the beat grid

One idea from the video made it into our own workflow right away. Before a single layer is built, the film exists as a beat grid: one row per event, every time locked to the music.

We don’t synthesize the track. We pick it from a real music library and measure it: tempo, every beat, the first beat of each bar, the strongest accents and the bars where the level jumps. Cuts and camera moves land on the first beat of a bar, small pops land on beats, the logo lands on the biggest hit. Here is the start of a grid at 100 BPM:

The client approves this table, not a render. Feedback like “hold the card longer” becomes one more bar in a row instead of a new build. Once the grid is approved, every keyframe time in the build comes from the measured beats, and the review checks that each cut sits within one frame of its downbeat.

Which one to use

If you need twenty product clips this month and nobody will ask for changes, write the harness and let the code approach render overnight. It’s cheap, and it gets better with every model.

If the film has a client, a brand team or a voice-over, keep it in After Effects. Let Claude build the first version with MotionUI, then finish it the way motion designers always have: by opening the comp.

The quickest way to see the difference is with one of your own Figma frames. Set up MotionUI with Claude Code and animate it.