Text-to-video has a control problem, and everyone who has used it knows exactly where it lives. You can describe a shot. You cannot really direct one. Where the camera starts, how far it travels, which lens it is on, when the actor crosses the frame, where the cut lands: all of that gets flattened into a paragraph and handed to a model that will make its own decisions about every single one.
Directors have never worked that way. Before the expensive day, there is a cheap one: grey blocks standing in for actors, a rough set, a camera moved around until the shot works. Previsualisation. You solve the shot when it costs nothing to change.
3D Director puts that day inside Kolbo.AI.
Grey First, Finished Second
You build the scene as a blockout. Primitives for the set, articulated characters for the people, lights where you want them. Nothing is textured and nothing is pretty, and that is the point. A blockout is for answering questions about space and timing, and it answers them in seconds.
Then you direct it. Dolly, truck, crane, orbit, pan, tilt, handheld. Lenses from 12mm to 135mm on a fixed virtual sensor, so a focal length means what it means on a real camera. Up to 20 shots and 30 seconds on a keyframe timeline, with a cut strip that shows real rendered thumbnails rather than guesses.
When the staging is right, you render it into a finished video.
That image is one shot, split. The left half is the blockout exactly as the tool renders it. The right half is what came back after it was sent to the video generator with the blockout itself as the camera and motion reference. The car is in the same place. The camera is on the same move. The road runs the same way. What changed is everything you were never trying to control by hand.
The Camera Is the Feature
The part worth dwelling on is not that you get a 3D scene. It is that the camera move survives.
A prompt that says "slow dolly past the car at golden hour" is a wish. A camera on a keyframed path, with a known focal length and a known duration, is an instruction. When that blockout goes to the generator as a reference, the finished shot inherits the staging you already approved instead of reinventing it.
This is why the grey pass matters. You are not building a 3D scene for its own sake. You are building the one thing a text prompt cannot carry.
What It Handles
Four blockouts, four finished shots. A character walking a night street, a quiet interior with motivated window light, a product turning on a pedestal, a figure entering through a doorway. Different staging problems, same workflow: block it grey, direct it, render it.
Kobi Builds It With You
You do not have to place every box yourself. Describe the scene and Kobi builds and edits it directly, as single undoable steps: reading the scene, positioning actors, setting camera coverage, adding motion, applying materials. Part way through it stops and looks at the shot, rendering a frame and checking it against what you asked for.
Two modes. Brainstorm plans without touching anything. Create makes the change. Every turn is undoable, so the scene is never somewhere you cannot get back from.
Getting It Out
Export the sequence as video, a single frame as an image, or the whole scene as GLB if you want to keep working in Blender or another 3D package. Everything lands in your Media Library, and there is a one-click handoff into the video generator for the finishing pass.
Honest Limits
3D Director is desktop only. It leans on browser APIs that mobile does not give us, and a timeline with 20 shots on it is not a phone interface. Video export needs a Chromium browser such as Chrome or Edge.
Character animation comes from a pose library and keyframes. It is previsualisation, not motion capture, and it is not arbitrary text-to-motion. It is built to answer where someone stands, when they move, and what the camera does about it.
The ceiling is 20 shots and 30 seconds per sequence.
Frequently Asked Questions
Do I need to know 3D software to use it?
No. The blockout is primitives, characters and lights, and Kobi will build and adjust the scene from a description. If you have used a timeline in any video editor, the sequencer will read as familiar.
Does the finished video actually follow my camera move?
Yes, that is the whole design. The blockout goes to the video generator as a reference for camera, layout and motion, so the finished shot inherits the staging you set rather than inventing its own.
Can I take the scene into Blender?
Yes. Export as GLB and the cameras, meshes and animated objects come with it.
How long can a sequence be?
Up to 20 shots and 30 seconds in total.
Does it work on my phone?
Not yet. 3D Director is desktop only, and video export needs a Chromium browser.
3D Director is live now. Open the 3D tools in Kolbo.AI, describe a scene, and direct it.