v0.2.1 · closed betaProfessional & scalable video production API/MCP/SDK

Give your AI agent a professional
video production engine.

Your agent writes the plan, previews the result, and renders the video.
A 15 second 4K clip in under 4 seconds, byte-identical on every rerun.
Built with Rust. GPU native. Timeline as state. No headless browser.

Request access →View docs
No credit card required · API + MCP · Preview before render
Launch teaser in FrameLane
Make a 12 s launch teaser from the rocket clip
Used 14 tools›
Launch teaser is saved as a project at v11: the rocket clip, a title, three caption plates and a ring mark top right. Its last preview came back with 0 violations.
Open it here so I can see it
framelane: get_project· Launch teaser · v11›
Opened Launch teaser in Browser›
It’s open on the right: 12 s, 1920×1080, 6 elements. What I change with edit_project shows up there live, and what you change there saves to the same project.
Type / for commands
Accept editsOpus 5.5
FrameLane: Video Editing & Rendering API for AI Agents
framelane.io/projects/proj_01M2Z0PCM0V37E3S3E28JT97KB
Launch day.
Now in beta
Timeline0:04.5 / 0:12.0100%Fit
0s1s2s3s4s5s6s7s8s9s10s11s12s
Ignition.
Now in beta
Same project, same engine.
Launch day.
ring-mark.png
clip7.mp4
Launch teaserSaved · v11ShareExport
VideoAudioImageTextShapesCaptionsAIBrandProperties
01 / EXAMPLES

What agents can make
with FrameLane.

Start with real footage or generate one. Let your agent edit and preview a project. FrameLane renders the final video.

USE CASE

Personalized at scale

AGENT PROMPT

“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.”

Templated projectPer-customer swapRecipe personalizationDynamic offersBatch renders

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.

proj_01HX5DGdone ● total 18.2s
02 / PIPELINE

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"
}
03 / ENGINE

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.

Rustmemory-safe core
wgpuGPU abstraction
WebGPUcheap validate + preview
VulkanLinux / GCP
GPU-accelerated effects on every platform
Preview runs the same engine as the final render
Edit, validate, and preview for free; pay only for a valid render
No "looked fine in the browser" surprises
Embed a WebGPU based video editor inside your application
INproject + ops{ edit → validate → preview }
FRAMELANE RENDERER
Rust · wgpu
↓↓↓↓↓↓↓↓↓
fragment shaders

Integrated with leading models

Kling V3Kling AI
Seedance 2.5ByteDance
Veo 3.1Google
Eleven v3Eleven Labs
and more
04 / WHY FRAMELANE

Built for real video,
not just motion graphics.

  • PerformanceBrowser overhead is real

    “10x more time and resources than bare FFmpeg because of the chromium layer.”

    ↗FrameLane has no browser in the render path: hardware NVDEC decode, one GPU compositor pass, hardware encode.

  • GPUNo GPU, slow frames

    “1-2 fullhd frame per second on server... quite expensive.”

    ↗FrameLane renders on GPU on every job: hardware decode, a GPU shader pass, hardware encode.

  • ScalingDoesn't scale with hardware

    “the render becomes slower as I increase the concurrency.”

    ↗FrameLane's GPU pipeline scales with hardware, not against it: parallel decode, one compositor pass, parallel encode.

  • QualityA quality ceiling in browsers

    “professional-quality video rendering is not currently possible in the browser environment.”

    ↗FrameLane never renders in a browser. It's the same native Rust/wgpu compositor in preview and in the final file.

  • DeterminismNot a deterministic renderer

    “you cannot do it perfectly because it's not a deterministic renderer.”

    ↗FrameLane renders the same timeline to the same bytes, every time.

  • TypographyTypography gets rebuilt

    “text needs to be rasterized to vector shapes and rendered as such.”

    ↗FrameLane renders type natively in the compositor, glyph by glyph, with no DOM or canvas text layer to work around.

  • ShotstackOpaque failures

    “the API could return an error to actually tell us why it's not rendering.”

    ↗FrameLane validates a project for free before it renders, and a failed render tells your agent why.

Capability
FrameLane
Remotion
Shotstack
FFmpeg
Timeline editing
Yes
limited
Yes
Yes
Manual
GPU shader effects
Yes
40+
limited
EC2 GPU only, not Lambda
No
Presets, no GLSL
No
Preview = render output
Yes
WASM, free before render
limited
CSS/WebCodecs drift
No
Separate engine
No
Stateful projects (edit, don't re-render)
Yes
Targeted ops + free validation
No
Regenerate each pass
No
Resubmit whole job
No
MCP / agent native
Yes
limited
Docs MCP only
Yes
No
4K output
Yes
limited
4K slow/costly
limited
1080p default, 4K enterprise
limited
Manual
No browser required
Yes
No
Core architecture
Yes
Yes
Managed (no cloud account)
Yes
No
Your AWS account
Yes
No
Built with
Rust/WGPU
React
CPU cloud
C
05 / FAQ

Common
questions.

Don't see your question? Email us →

Companies shipping video as a product feature, AI agents, or pipelines producing high-quality video at scale.
Your agent patches the timeline and previews any frame before it renders anything. A change is a targeted, verified operation and not a complete rerun. Fix a caption's timing, swap a graphic, adjust an animation, you change that one thing and not the whole video.
Remotion renders a browser page and screenshots it, which is why 4K export is slow and why frames drift. Template APIs give you slots in a fixed layout. FrameLane is a real timeline with a GPU compositor underneath: layouts reflow instead of cropping, output is identical byte-for-byte across runs, and nothing in the render path is a browser.
After Effects farms give you AE fidelity at AE speed — minutes per clip, CPU-bound, and you need someone who builds AE projects. FrameLane renders its own compositor on a GPU, so the same job takes seconds and the template is a data structure rather than a project file.
Seconds per clip at 4K, because it renders on GPUs rather than in a browser. Free covers 25 render-minutes at 1080p with a watermark. Pro is $99 a month for 2,000 render-minutes at 4K, then $0.08 per render-minute after that. A ten-second video is a sixth of a render-minute.
Yes. Free tier, self-serve API key, first render in minutes. The founder answers every signup personally within a day, but nobody makes you sit through a call.
You give us one design, or one existing video you want more of. That becomes a template with the parts that change marked as inputs — a name, a language, a number, a product shot. After that every render comes back with your type, colours and layout intact. During beta we build the first template with you.
Either. Bring your own clips, stills and voice, or have FrameLane generate them — it calls the current frontier video and voice models on your behalf. Generated material is stored as an asset the moment it's made, and your timeline references the asset. The clip is generated once; the video renders identically every time after that.
06 / GET STARTED

Get started
in seconds.

Install the SDK and start building with FrameLane: edit a project, preview it cheaply, and render when it's valid.