What agents can make
with FrameLane.
Start with real footage. Let your agent edit and preview a project. FrameLane renders the final video.
Personalized at scale
“Pull this week's loyalty record per customer — first name, top-matched recipe, and their two best savings offers — drop them into the fixed Waitrose template on the text and image layers, keep every transition, price-tag reveal, and pacing cue identical, then queue a render per customer record.”
One project, sent a thousand ways: each customer's name, recommended recipe, and savings offers change on every render while the branding, transitions, and pacing stay identical, driven by a single API call per customer record.
Everything an agent needs
to produce video.
A complete editing and rendering pipeline exposed through API, SDKs, and MCP.
Asset ingest
Reference media by URL or upload it, then point a video color grade at a .cube LUT. SVG assets rasterize in process, so a stroke icon can be a sticker or a watermark with no PNG step. FrameLane stores each asset and probes its metadata. Video sources must be MP4, MOV, or WebM.
{
"source_url": "https://cdn.example.com/clip.mp4"
}How FrameLane
works.
The rendering engine is written in Rust and built on wgpu. The same codebase compiles to a native binary that runs on Vulkan on GCP GPU instances, and to WebAssembly that runs against the browser's GPU API in your web application.
Integrated with leading models
Built for real video,
not just motion graphics.
- GPUNo GPU on Lambda
Lambda and Vercel Sandbox do not have GPUs. You are limited to CPU-only rendering.
↗FrameLane renders on GPU. Every job.
- RenderingCSS animations are forbidden
CSS transitions or animations are FORBIDDEN, they will not render correctly.
↗FrameLane is GPU-native. No browser. No frame simulation. No flickering.
- AI AgentsAgents have to write React
No React, no proprietary DSL. AI-first, agents already speak HTML/JSON.
↗FrameLane is JSON-in, video-out. Any agent, any language, one API call.
- TimelineThe edit layer is a black box
Agents can write scripts and captions, but they cannot read or modify the edit, so every pass is a regeneration, not a refinement.
↗FrameLane exposes the edit as a stateful JSON project your agent reads and patches with targeted ops: tracks, clips, captions, timing, validated for free before it renders.
- PerformanceGrinds to a crawl with real video
After about 20-30% of rendering 4k video is complete, it grinds to a crawl and renders at about 5fps and continues to slow.
↗GPU decode, GPU encode, parallel rendering.
Common
questions.
Don't see your question? Email us →
Get started
in seconds.
Install the SDK and start building with FrameLane: edit a project, preview it cheaply, and render when it's valid.