Remotion vs Motion Canvas

Two Mental Models

Remotion's model is declarative. A composition is a React component that reads the current frame and the video's configuration and returns what should be on screen at that frame. Animation is a function of time: interpolate a value between frame 0 and frame 30 and the component renders accordingly. That makes any frame independently renderable, which is what allows Remotion to split a render across many machines, and it means everything a React developer already knows (components, props, CSS, data fetching) carries over.

Motion Canvas's model is procedural. A scene is a generator function that creates objects and then yields animations in sequence: move this, then fade that, then wait for the narrator. The editor shows a timeline of those steps with a real-time preview and lets you place time events against an audio track. For explainer videos, where animation follows a script and a voice-over, this reads more naturally than frame arithmetic. It is a better fit for what it was designed for, and a worse fit for open-ended layouts or video built from data.

Rendering and Infrastructure

Remotion renders by driving a headless Chrome build it installs itself, capturing each frame and encoding the result. Because rendering is a server-side operation, Remotion has built an infrastructure story around it: distributed rendering on AWS Lambda (its recommended option), Vercel Sandbox, plain Node.js or Bun APIs, and Cloud Run (which its docs describe as alpha and not actively developed). Since v4.0.491 it can also render in the browser with WebCodecs, for a subset of HTML elements. The @remotion/player component embeds compositions in any React app with inputs set at runtime, which is how many products offer in-browser previews before a server render.

Motion Canvas renders inside its editor: it plays the timeline and saves each frame as PNG, JPEG or WebP, and an FFmpeg-based exporter, which its docs call still relatively new, can write video with audio. There is no built-in headless or cloud rendering, which is the main gap the forks set out to fill.

Maintenance and the Fork Question

This is the deciding factor for many teams. The Motion Canvas core repository's last release was 3.17.2 on December 14, 2024, followed by a 3.18.0 alpha in February 2025, and its last code commit was in February 2025. A community pull request opened in September 2026 proposes modernizing the build, test and docs toolchain, explicitly without new features. That's useful, but it isn't a roadmap.

Two forks carry the work forward. Canvas Commons (MIT), a community-based fork created in July 2025, released 0.4.0 on September 24, 2026, adding text splitting and text on paths, justified wrapping, hand-drawn shapes, animated layouts and faster export. Migrating from Motion Canvas is mostly a package rename. Revideo (MIT), now maintained at github.com/midrender/revideo, released 0.11.0 in July 2026 and adds headless rendering, a render endpoint and CLI, Cloud Run deployment, parallelized rendering, a React player and audio and video export. Note that it sends anonymous render telemetry by default unless disabled. Remotion, by contrast, ships releases continuously and is maintained by the company that sells its license.

AI Agents and LLM-Written Animation

Language models now write a large share of motion-graphics code, and Remotion has leaned into that. It publishes 12 official Agent Skills (installed with npx skills add remotion-dev/skills) that teach Claude Code, Codex, Cursor and others how to write correct Remotion code, plus "Prompt to Video" and "Prompt to Motion Graphics" starter templates. React's huge presence in training data helps too: models already know the idioms. See SKILL.md for how skills work.

No official agent tooling was found for Motion Canvas or its forks. Models can still write generator-based scenes, but with less help and a smaller body of examples. Either way, running AI-written code in a render pipeline is a security question as well as a quality one. Code that renders in a browser can fetch from the network and import arbitrary packages, so pipelines that accept generated components should sandbox the renderer, restrict imports and require human approval of the exact code that runs (see agent sandboxes).

Licensing and Cost

Motion Canvas and both forks are MIT-licensed, with no conditions on company size or use. Remotion uses its own source-available license. Individuals, for-profit companies with up to three employees, non-profits and evaluation don't need a company license; companies of four or more do. As of October 2026 the company license is $25 a month per seat for Creators, $0.01 per render with a $100 monthly minimum for Automators (rendering as a service), or Enterprise from $500 a month. The license doesn't allow reselling derivatives, and a proposed Remotion 5.0 change would count contractors toward team size. For most businesses the cost is small next to engineering time, but it is a real procurement step that MIT-licensed tools don't require.

Which Fits Your Pipeline

The practical split is by output. If you are generating many videos from data, embedding previews in a web product, rendering in the cloud or having agents write the code, Remotion's ecosystem is far ahead. If you are hand-crafting explainer animations synchronized to narration and want a fully permissive license, Motion Canvas's model is elegant, but start with Canvas Commons or Revideo rather than the dormant core. For how either fits into a wider edit, see video as code vs nonlinear editing.