Skip to content

Why framefields works well with coding agents

View .mdllms.txt

A coding agent can write a video in TypeScript, render a few frames, inspect the results and adjust the composition. Framefields gives it the tools to do that without opening a desktop editor.

The engine renders compositions with WebGPU in Node.js. You describe layers, animation and timing in code, then render a single frame or an MP4. The same source can be read and edited by a person or an agent.

A full video takes longer to render than a still. Framefields lets an agent work through smaller checks first:

  1. Type-check the project and run comp.lint() to find common composition problems.
  2. Render a frame at a useful moment and inspect it for layout, text and color.
  3. Render a frame grid to check a sequence of moments.
  4. Open the live preview for playback, scrubbing and notes.
  5. Export the video, with optional QA checks.

The rendering guide shows the commands and code for each step. Frame grids are useful for ordinary animation, but not for render-time vision: that needs consecutive frames because its results lag by one frame.

A human can pause the preview and leave a note on the picture, such as “move the title up.” The note is saved locally, so an agent can read it and update the composition. You’ll still want to check the result together. A successful render doesn’t automatically mean the timing feels right.

Framefields includes text animation, 2D layers, 3D models, effects, audio and vision. You can see the code and output in our examples:

The engine is also behind Gatewai Studio. Read more about its background.

For Claude Code:

/plugin install framefields

For other coding agents that support the skills installer:

Terminal window
npx skills add gatewai-dev/framefields

Then give the agent a specific brief, including the size, duration and moments you want to inspect. For example:

Create a five-second product teaser at 1080x1920 and 30 fps.
Add an animated headline and a bar chart. Render frames at the
start, middle and end so I can check the layout before export.

Framefields needs Node.js, project assets such as fonts, and a suitable rendering environment. CPU rendering is available on supported setups but is slower. See getting started for requirements and the agent guide for the workflow.