Catch
Character and wardrobe drift against your references, plus black frames, freezes and flicker. Every problem comes with the frame, the time and one suggested fix.
Kolam Labs · Applied AI for film and video
AI that makes generated video dependable.
Video models can now turn a sentence into a scene. We build the intelligence that makes what they generate consistent, verifiable and ready to ship, at any scale.
Raccord Check
Is this take good?
The quality gate for AI-generated video. It checks every take from any model, says what is wrong in plain words, and stops bad renders before they cost money.
Request early access
Launch expected Q4 2026
Raccordtake_02.mp4
Fail
Don’t use this take: 2 problems found.
Brightness flickers in the upper right of the frame from 0:01.7 to 0:02.8.
Fix: re-render with a different seed.
The picture freezes for 0.6 s at 0:04.0.
Fix: re-render with a different seed.
Why it matters
Every second is billed. Many takes are thrown away. Someone has to watch them all.
$0.40
per second of video for Veo 3.1 with audio, at Google Cloud list price.1
300–400 → 15
clips generated vs clips used for one AI-made TV ad, Kalshi’s 2025 NBA Finals spot.2
70,000+
clips sorted by a team of five in 30 days for Coca-Cola’s 2025 AI holiday ad.3
Models now try to keep characters consistent. On the vendor pages we reviewed, none checks the finished take against your references.4 So a person still does.
- Google Cloud, Vertex AI generative AI pricing, Veo 3.1 with audio, 720p/1080p, as listed November 2026. cloud.google.com/vertex-ai/generative-ai/pricing
- The Keyword, 16 June 2025, on Kalshi’s NBA Finals ad made with Veo 3. A single production, not an industry average. thekeyword.co
- TheWrap, 3 November 2025, on Coca-Cola’s 2025 AI holiday ad. thewrap.com
- Consistency features reviewed November 2026: Runway Gen-4 (runway.com), Kling Element Library (kling.ai), Veo 3.1 Ingredients to Video (blog.google), LTX Elements (ltx.io). “None we reviewed” is not a claim that none exists.
What it does
One question. One answer. For every take.
Rank
Give it a batch of takes. It puts the best two in front of you, so you watch fewer clips.
Stop early
Checks the start frame before you pay for motion, stops asking for takes once one passes, and halts drifting renders on open models.
Where it fits
At every point where money is about to be spent on a take.
- Before renderCheck the prompt, the references and the start frame.
- During renderStop a take that is drifting, on models you run yourself.
- After renderA verdict per take, the best of a batch.
- At deliveryA signed record of what was checked.
Built to read output from any model
- Veo
- Kling
- Runway
- Seedance
- LTX
- Wan
- Hailuo
- and the next one
It never generates video itself. Models come and go; your quality gate shouldn’t. Model names are trademarks of their owners, used only to describe compatibility.
Built for pipelines and agents
Wherever generated video lands, it can be checked.
A paid API priced per second checked, with an MCP server so AI agents can call it directly, a ComfyUI node, watch folders and a command line for pipelines.
- REST API
- MCP server for AI agents
- ComfyUI node
- Watch folders
- Command line
- Review screen
How we build it
Honest by design.
Plain words
Every problem is one sentence a producer understands, with the time it happens.
Honest confidence
When it isn’t sure, it says “not sure, take a look” instead of guessing.
Signed verdicts
Each verdict records what was checked and by which version, and can be re-run.
Private by default
Your clips are deleted after checking unless you choose to keep them, and never used for training without your consent.
We publish our accuracy and how we measured it before we ask anyone to rely on it.
SceneRight
Direct the scene. Don’t prompt it.
A complete directing environment for AI film. Cast a company of characters, build the sets, stage and cover the scene, cut it to a timed reel, and watch the episode before a single frame is rendered.
Talk to us about SceneRight
Launch expected Q1 2027
Cast once
Characters and sets are assets, not descriptions, so every shot holds continuity.
Stage and block
Place the actors, call the lens and the move, shape the performance.
Cover and cut
Wide, singles and inserts, cut to a timed reel with dialogue and music.
Render last
Watch the episode first, then render what was directed through the model that suits it, on your own keys.
Get the scene right before you spend a rupee or a dollar rendering it.
For anyone making serialised video at volume: microdrama platforms, serial studios, agencies and animation houses. Operable by people who have never written a prompt.
About Kolam Labs
One line, many points, nothing out of place.
Kolams are drawn at the threshold of homes across South India. In one kind, the sikku kolam, a single continuous line loops around a grid of dots without a break. Our name is a reminder of the craft we care about: making many pieces hold together as one.
Kolam Labs LLP is based in India and builds for productions everywhere. We are building two products: Raccord Check, the quality gate for every generated take, and SceneRight, the place where the scene is directed before anything is rendered.
Our vision
A world where anyone can make film with AI, at any scale, and trust every frame they ship.
Our mission
To make AI-generated film and video dependable enough to build on, for every studio, platform and agent that makes it.
What we believe
01
Confidence is the new frontier.
Generation is getting cheaper and better every quarter. What stays scarce is knowing that what came out is right. That is where we push.
02
Quality must be neutral.
Models launch, improve and get retired. A standard for what is good has to hold across all of them, and outlive any one of them.
03
Machines will need judges.
As AI agents and pipelines make video at machine volume, checking has to run at machine speed, with answers a person can still read.
We work where computer vision meets generative AI, and we measure everything we claim.
Make every take count.
Studios, platforms and teams generating video at volume: we would like to hear from you.
hello@kolamlabs.ai