21 camera moves
Push in, orbit, dolly, crane, low hero, close-up, two-shot, over the shoulder, pan between, follow behind, fly alongside, handheld track. Named first, editable by hand after.
Text prompts are bad at camera work. Describe your scene here and you get a simple 3D blockout with a real camera move through it — then export a short reference clip that shows a video model exactly how to frame and move.
Free scenes to start, no card. The studio itself stays free forever.
“A dragon flies low over a village, banks around the tower and lands in the square while villagers scatter.” You get blocks, motion and a shot list.
Drag anything. Scrub the timeline and drag again to lay a path. Blocks are the point — nobody is modelling a dragon here.
Pick a move — push in, crane down, over the shoulder, fly alongside — then drag the camera and its aim in the viewport until the framing is right.
One clip per shot, plus a prompt written to match it. Give both to your video model and the camera does what you asked.
Push in, orbit, dolly, crane, low hero, close-up, two-shot, over the shoulder, pan between, follow behind, fly alongside, handheld track. Named first, editable by hand after.
One timeline for the action, another for the edit. Shots are windows onto the action, so cuts stay continuous and two cameras can cover the same moment.
Pin any camera key to an exact second, or drag it along the timing strip. Hold a beat, then move.
Export the same shot as colour, depth, normals or line art — the inputs structure-conditioned models actually want.
Every shot writes its own prompt from the real framing: lens in millimetres, camera height, what is moving in frame.
Clips encode offline, several times faster than real time, straight to MP4. Your work autosaves as you go.
The studio is free and stays free — blocking, cameras, motion, terrain, timing and export all run in your browser and cost us nothing to give you. You only pay for the AI that writes a scene from a sentence.
Freeno card
Scene generation on Claude Sonnet.
Freeforever
When the free scenes run out, generation is all that stops.
—one-off, no subscription
Buy scenes when you need them. They do not expire.
A generation that fails is refunded automatically, and a camera tweak costs a fraction of a full scene — you are metered on what the model actually did, not on a flat per-click fee. No subscription, and nothing is charged until you choose a pack. Refund policy.
Previsualization, or previs, is a rough animated version of a scene made before it is shot, used to work out camera positions, movement and timing. It is deliberately unfinished: grey placeholder shapes stand in for sets and actors so the decisions are about framing and pacing rather than detail.
Because camera direction is ambiguous in text. "Push in slowly, then whip pan to her" describes a move that could be executed a hundred ways, and the model picks one. Showing the move in a reference clip removes the ambiguity, because the motion is in the pixels rather than in the words.
Give it a reference clip instead of describing the move. Block the scene out roughly, move a virtual camera through it the way you want, export a few seconds of that, and pass it to the model as a motion or video reference. The model then has the framing, the speed and the direction to follow.
No. You describe the scene in a sentence and get a blockout back, then adjust it by dragging. The shapes are boxes, cylinders and blocked-out figures rather than models, so there is nothing to sculpt, texture or light.
Grey placeholder geometry standing in for people, vehicles and buildings, animated on a shared timeline, filmed by one or more virtual cameras with real focal lengths and moves. What it carries is spatial: where things are, how they move, and how the camera covers them.
A new account gets 6 free scene generations with no card and no time limit. After that, credit packs start at $9 for 25 scenes. Everything except generation - blocking, cameras, motion, timing and export - stays free forever.
Scenes are saved to your account so they follow you between browsers, and your prompt is sent to the AI provider that answers it. Rendering and video encoding happen in your browser, so exported clips are never uploaded.
Yes. A scene has one action timeline and any number of shots looking at it, so several cameras can cover the same moment from different angles. You can cut between them and export each angle separately, which is usually what a video model wants.