Skip to main content

​
What is Framelane?

Framelane exposes your video production pipeline as a REST API. Submit a composition, get back a rendered video. Run transcription as a standalone async task when you need text out of audio or video. Every operation follows the same pattern:
Wait using either:
  • Polling — GET /v1/renders/{id} or GET /v1/tasks/{id} until status is completed or failed
  • Webhooks — subscribe to render.completed / task.completed (recommended for production)
Bring media into Framelane storage when needed:
  • Local file — POST /v1/uploads
  • Remote URL you don’t control (a direct public link to the file) — pass the URL as an element source_url; the render engine fetches it at render time, and video/audio are probed for duration first unless you set out_point
  • Your own CDN — pass the URL directly as source_url (no upload or copy)
The Framelane Renderer is a GPU-accelerated native compositor — not FFmpeg. Compositing, effects, and transitions run on the GPU, which is what makes complex multi-layer timelines cheap; encoding is constant-quality, MP4 on the GPU encoder at the renderer’s default quality, WebM and MOV on the CPU. Renders with motion_blur re-render every frame samples times with CPU-side accumulation — expect roughly samples× the render time.

​
Key Concepts

Renders

Compose and render video from a timeline of five element types — video, audio, text, image and vector shape (SVG path, hex-or-gradient fill, stroke, dash, morph, and draw-on trim) — plus transitions, transform groups with stack layout and nesting, and custom_animations: inline keyframe registries for motion the preset catalog doesn’t name.

Capabilities

GET /v1/capabilities — unauthenticated. The machine-readable catalog every agent should read first: flat enum lists, element_schemas (per-element JSON Schema), and request_schema (groups, custom_animations, motion_blur, background_gradient, transitions).

Preview

Validate a composition for free, or render a cheap frame, clip, or whole-timeline contact sheet before spending a full render.

Tasks

Standalone AI jobs: transcribe audio or video to text with word-level timestamps.

Webhooks

Subscribe to job lifecycle events. HMAC-signed payloads with a 7-attempt retry schedule and dead-letter queue.

Guides

How-tos for the REST API: author a composition, match a reference, render, transcribe, cut a captioned social clip, batch tasks and set up webhooks.

Agent Skills

Workflow skills for AI agents, served by the MCP server as prompts and as get_capabilities sections.

Playground

Build a composition and preview it in the browser with WebGPU — the same renderer as the API — then copy the exact render request.

​
API Base URL

All endpoints use HTTPS. All request and response bodies are JSON.