# Compared with other tools Other ways to make videos with code and agents, what each is good for, and where Intentface Launch fits among them. Intentface Launch is an exploration. We built it for our own launch videos and keep developing it as we explore programmatic video. It is made to be shaped by product teams around their own product, brand and way of working, more a harness to adapt than a finished tool. It is MIT, and we do not sell it, host it or have a paid tier, so we have no stake in which tool you pick. This page is a map of good options, each described on its own terms. Most are free to try: make the same short video in two or three of them and see what works best for you. Checked on 18 October 2026 (cinetic on 19 October) against each project's own README, docs and package registry. These projects move fast; if something here is out of date, please open an issue. ## Where to start A starting point by what you are making, not a ranking. Several of these combine well, and many teams will use more than one. - **Videos from a website URL, a script or a brief, including explainers, captions, talking-head recuts and decks, across many coding agents, with cloud rendering:** [HyperFrames](#hyperframes). - **A general React framework for video, for production work, with an embeddable player and serverless rendering:** [Remotion](#remotion). - **Polished product-demo shots on Remotion (chat streams, terminals, device frames), themed like shadcn/ui:** [snapcn](#snapcn). - **Diagrams and technical explainers drawn on a canvas:** [Motion Canvas or Revideo](#motion-canvas-and-revideo). - **Maths and algorithms, in Python:** [Manim](#manim). - **A launch video built from your product's own source and running UI, adapted to your team's brand and workflow, with Claude Code on a Mac:** [Intentface Launch](#intentface-launch). ## At a glance | | Intentface Launch | HyperFrames | Remotion | cinetic | snapcn | Motion Canvas, Revideo | | --- | --- | --- | --- | --- | --- | --- | | **You write** | React shots | HTML, CSS, GSAP and the like | React | Remotion or HyperFrames, written by the agent | Remotion components | TypeScript scenes | | **Licence** | MIT | Apache-2.0 | Free for individuals and small teams; a company licence above 3 employees | MIT (the engine's licence applies) | MIT, plus paid Pro | MIT | | **Agents** | Claude Code | Most coding agents | Official skills | Most coding agents | Skills, MCP with Pro | None official | | **Your UI** | Imported and captured | Captured from a URL | Built or imported | Rebuilt | Recreated | Drawn | | **Cloud rendering** | No | Yes | Yes | No | Through Remotion | Revideo's render API | | **Version** | 0.2 | 0.8 | 4.x | 1.1 | Months old | 3.17, 0.11 | All of them render deterministically when scenes follow their rules (time from the frame number, seeded randomness): the same input gives the same frames. Each section below has the details and links to the licence and pricing pages. ## Intentface Launch A React video engine, a CLI, a local studio and a set of Claude Code skills, built around one job: a launch video of a software product, made from the product itself. Claude reads and runs your app, agrees a brief and script with you, cuts to a music track's beats, sketches a storyboard you comment on, then builds and renders shot by shot. **Good for** - **The real product.** Videos import your components and data. Capture scripts drive the running app one exact frame at a time, and DOM and CSS snapshots keep close-ups sharp at any zoom. - **Shaping it to a team.** Core is brand-neutral. A brand add-on carries a team's fonts, logos, design, and motion and copy rules. `launch/CLAUDE.md` holds how your team works, and rules written there after one video carry into the next. - **Review before building.** You comment per shot on a storyboard and animatic before anything heavy is built. - **Checks a person usually does by eye.** `pnpm launch check` measures every line of text on screen for reading time, size, contrast and room from the UI beside it, and flags end frames that differ from the approved logo. - **Craft add-ons.** 3D with live UI on panels and real depth of field, and WebGPU shaders on the video's clock. - **MIT throughout**, with no Remotion dependency and no telemetry. **Know before you choose it** - **Young.** On npm since 13 October 2026; 0.1.1 and 0.2.0 shipped broken installs, fixed in 0.2.1. APIs, commands and skills will change. - **Tested on one setup.** An Apple M2 Pro Mac with Claude Opus 5.5 at High effort. Other operating systems, GPUs and models are untested, and without a GPU rendering is much slower. See [Requirements](/requirements). - **Claude Code in practice.** The skills install in other agents through the skills CLI, but we have not tested that. - **Built for your source and a running app.** Captures can record any URL, but the method expects the product's source and a demo or seeded mode you control. - **Narrow.** Product launch videos, not explainers, captions, decks or general motion graphics. - **No hosted rendering**, and no editor beyond the review studio. - **No music.** You bring or generate a track; the video is cut to it. ## HyperFrames [HyperFrames](https://hyperframes.dev), from HeyGen, renders HTML, CSS, media and seekable animations to deterministic MP4: "Write HTML. Render video. Built for agents." It is used in production at HeyGen, Apache-2.0 ([licence](https://github.com/heygen-com/hyperframes/blob/main/LICENSE)), and moving fast. **Good for** a broad range of videos made by an agent: launch videos from a product URL, script or brief (its `/product-launch-video` skill, which shares a name with ours), faceless explainers, PR-to-video, captions, talking-head recuts, motion graphics and music videos. It renders in headless Chrome with FFmpeg, its 21 skills come as plugins for Claude Code, Codex, Cursor, Gemini CLI and GitHub Copilot, it captures a site's visuals and brand from a URL, and it renders locally or in HeyGen's cloud, on AWS Lambda or on Google Cloud Run. There is also a Studio app for working without code. The breadth is impressive. **Know before you choose it:** it is HTML-first, with no React runtime. A skill ports Remotion projects to HTML and GSAP, one way, so a product's own React components do not drop in. Cloud renders and some voice and music features use a HeyGen account. [Docs](https://hyperframes.heygen.com/introduction). ## Remotion [Remotion](https://www.remotion.dev) is the established React framework for programmatic video, and a solid choice for production work: Studio, an embeddable Player, rendering on AWS Lambda, on Vercel or in the browser, a large ecosystem and [official agent skills](https://www.remotion.dev/docs/ai/skills). It has been in production for years, and much of programmatic video, ours included, builds on ideas it made common. **Good for** React video of any kind: video inside your own app, templated videos at scale, and anything that benefits from a mature ecosystem and documentation. If you know Remotion, our engine will feel familiar (`useFrame` for `useCurrentFrame`, `Sequence`, `interpolate`, `spring`). **Know before you choose it:** it is free for individuals, non-profits and companies of up to 3 employees, and a for-profit company above that needs a company licence, which funds the project. Remotion 5.0 (announced, not yet released) changes the terms slightly, for example counting contractors towards team size. Read the current [licence](https://www.remotion.dev/docs/license) and [pricing](https://www.remotion.dev/docs/license/pricing). ## cinetic [cinetic](https://github.com/Leonxlnx/cinetic), by Leonxlnx, is an agent skill (MIT) that has your coding agent "direct a short film instead of arranging animation components": launch films, feature loops, logo stings, vertical videos and UI walkthroughs, built on Remotion by default or on HyperFrames. It installs with `npx skills add Leonxlnx/cinetic` or as a Claude Code plugin. First public release 28 September 2026; 1.1.0 on 18 October. **Good for** a complete, finished-looking film from a short brief, including an invented brand when there is none. The agent writes three concepts and picks one, draws its moves at random from a library of 273 measured techniques in 14 categories (so two films do not share the same push-in and end card), times everything on one beat grid, scores the sound with [earshot](https://github.com/Leonxlnx/earshot) on the frames the picture computes, and renders true motion blur with BT.709 colour. Then it reviews its own render with contact sheets, pixel forensics, an audio-sync audit and critic prompts. Its evaluation against the same agent without it is published round by round, losses included. Its technique library, each move with its timing in frames and when to use it, is worth reading whatever you use. **Know before you choose it:** it has a strong house style. When the agent invents the look, hard bans rule out serif and italic type, glow, glass, purple, orange and multi-hue gradients, and overshoot on copy; a brand you supply overrides them. Product UI is rebuilt from your components or by hand, not captured from the running app. Its own evaluation puts a run at about three times the time and tokens of the agent without it (95 minutes and 626k tokens on average). Remotion's licence applies when you build on Remotion. The scripts need Python with numpy, scipy, librosa and OpenCV besides Node and ffmpeg. Its description asks the agent to use it for any video request, so install it in a repo of its own rather than beside another video skill set. ## snapcn [snapcn](https://snapcn.dev), by Sri Nath under the community org Shadcn Labs, is a shadcn-style registry of Remotion components for product demo videos: AI chat streams, terminal sessions, device frames, captions, logo stings and whole scenes. You install components with `npx shadcn add`, and they are copied into your repo as source you own. It is a few months old and very active. We really dig shadcn/ui, and snapcn's approach is a cool one. **Good for** polished, recreated product-demo shots in a Remotion project, themed with shadcn/ui's tokens. It ships skills (including a launch-video workflow), a Claude Code plugin, an MCP server with Pro, and a browser editor that exports MP4, so you can try it without installing anything. **Know before you choose it:** it needs a Remotion project, and Remotion's licence applies. Free components are MIT; Pro components and the MCP server are paid, and company work needs its Commercial plan ([pricing](https://snapcn.dev/docs/pricing)). Shots are recreations of a UI rather than captures of yours, which also means they need no running app. Captured UI and recreated shots can sit in one film. ## Motion Canvas and Revideo [Motion Canvas](https://motion-canvas.io) animates on a canvas with TypeScript generator functions, with a web editor to sync scenes to audio. It is MIT and popular for explainers; its last stable release was 3.17.2, in December 2024. [Revideo](https://github.com/midrender/revideo), forked from it and MIT, adds a headless render API and a React player, and is the engine behind [Midrender](https://midrender.com). **Good for** diagrams, technical explainers and videos generated in batches, where the visuals are drawn rather than taken from a web app. **Know before you choose it:** scenes are drawn on a canvas, not DOM, so a web product's components do not drop in. Revideo sends anonymous render telemetry by default (`DISABLE_TELEMETRY=true` turns it off). ## Manim [Manim](https://www.manim.community) (the community edition, MIT) animates maths and algorithms in Python, and is the standard there. It is not aimed at product videos. ## An agent with no framework Ask a coding agent for a launch video with nothing installed, and it will usually write a script that screenshots pages with Playwright or Puppeteer and pipes them into ffmpeg. **Good for** a quick one-off or a proof of concept: you own every line and there is nothing to learn. **Know before you choose it:** each video rebuilds the same plumbing: deterministic timing, parallel rendering, audio and beat sync, transitions, legibility checks, safe zones for vertical cuts. That plumbing is what every tool on this page provides. ## Generated footage Video models such as [Veo](https://deepmind.google/models/veo/), [Runway](https://runway.com) and [Kling](https://kling.ai) make footage from a prompt, and they can start from images, including screenshots of your UI. Combining generated footage with launch materials is an avenue we are exploring. Video models are still expensive and heavy to run, so they are not part of the harness yet. A clip you generate yourself goes in a video's `media/`, with its licence recorded, and plays in a shot with `Clip`, beside the product's real UI. The same works in the code-based tools above.