FACEHUB AI
ServicesUse CasesPricing
Sign In
Scene Designer

Cuts That Survive the Generator

Lay out the room, place a camera for each shot, and cut between them. What reaches the video model is a recording of your blocking — so the cuts land where you put them and the characters keep looking at each other across them.
Filmmakers
Music Video Directors
Ad Creatives
Storyboard Artists
Animators
Try It Free View Pricing

A cafe scene blocked in 3D, four camera setups, and the generated result — 65 seconds.

The Problem

AI video models give you one continuous take and no say in it. Ask for a wide that cuts to a close-up and you get a slow drift through the room instead, because a model with no memory between shots smooths everything into one move. Characters look the wrong way, or straight into the lens. The scene you had in your head never arrives, and the twentieth regeneration is no closer than the first.

The Solution

Build the scene once as grey blocks: the room, the furniture, the people. Put a camera where you want it and lock it off. Move the playhead, frame the next angle, lock that one too. The whole thing is recorded frame by frame at a fixed rate and handed to the video model as a motion reference, alongside a written shot list that says exactly where the cuts fall. Geometry carries the blocking; words carry the performance.

How It Works

1
Describe the scene, or show a photo

Write it in plain English and the room, the furniture and the cast are laid out for you. Drag anything that is not where you want it, or add pieces by hand from the shape library.

2
Compose a shot, then cut

Fly the camera to a framing and press New shot — it locks off there and holds until the next one. Move the playhead, frame the next angle, press it again. Key the action on the timeline: who moves, who turns, who is looking at whom.

3
Direct it, then generate

Write how each beat is played — physical description of what the faces do. Attach reference photos for faces and location. The blocking is recorded as a clip and sent with a shot list naming every cut.

What You Get

Hard cuts between shots instead of one drifting move

Locked-off angles that genuinely hold still

Eyelines that match across a cut, because the geometry says so

Timed beats — say what happens at 0-4s and have it honoured

Direct facial performance in words, beat by beat

Reference photos for faces and location carry through

Download the reference clip and check it before you spend a generation

No 3D modelling, rigging or camera-rig software to learn

Example Results

FaceHub Scene Designer with a cafe laid out in grey blocks, two characters placed near the counter and camera keyframes marked along the timeline

The room, blocked out. None of this needs to look good.

A camera view inside the blocked cafe scene, grey placeholder figures seated at tables with one character highlighted in red

A camera placed for one shot, framing what it will see.

Two placeholder characters facing each other in the Scene Designer, one selected in green, with the shot timeline below

Two characters in one scene, so the eyelines are a fact.

Ready to Try It?

Start with 50 free credits. No credit card required.

Get Started Free

Frequently Asked Questions

It can, but only if you tell it where the cuts are and show it a reference that cuts. Left to itself a video model smooths a multi-shot request into one continuous move, because each generation has no memory of the last. Here the reference clip cuts, and the prompt names every cut time.

A shot is locked off by default. Each camera you place holds its framing until the next one, so nothing moves unless you ask for it. If you want the camera to travel into a shot you turn that on deliberately for that shot alone.

Because they are real objects in one 3D scene, so the direction each one faces is the same fact in every shot. The cameras are also placed off the eyeline, and on opposite sides for a pair — two people cut together must look in opposing screen directions or they read as looking the same way rather than at each other.

Usually because the camera was placed on the eyeline: if the person they are looking at sits directly behind the lens, a correct look is a look into the lens. Putting the camera 40 to 60 degrees off the eyeline sends the look across the frame instead.

No. The scene is built for you from a description or a photo, and you move things by dragging. There is no modelling, texturing or rigging — the 3D exists to carry blocking and camera, and everything you see is grey placeholder geometry.

Up to 15 seconds per generated clip, which is the limit of the underlying video models. Fifteen seconds holds four or five separate shots — enough for a complete beat of a story.