AI Rotoscoping: Mask a Moving Subject

Isolate a moving subject frame by frame. No hand-drawn masks, no green screen.

Every video model

One workspace

Mark a subject, get a mask

Rotoscope a clip in three steps

  1. 01

    Open your clip

    Drop a generated or uploaded clip onto the timeline and park the playhead where the subject is clearest.

  2. 02

    Mark the subject

    Click it, drag a box around it, or describe it in a few words.

  3. 03

    Use the mask

    Grade inside it, blur what it covers, or invert it and change the background.

Use cases

Grade one subject, not the whole shot

A take is right except for the subject: too cool, too flat, lost against what is behind. Track them and the grade lands on those pixels alone, so the rest of the frame keeps the look it was rendered or shot with. The alternative is regenerating a take you already like.

Pick the person you mean

Two people share a frame and only one needs the work. A rectangle can say where but never which, and a hand-drawn shape has to be redrawn every few frames. A click plus a short description settles it, and each person ends up with a mask of their own.

Lift a subject onto a new background

Invert the tracked mask and the subject comes away from its plate, ready to sit over another clip, a gradient or a flat colour. No green screen at the shoot, and no keyer struggling with a cluttered room, because the model reads the picture rather than a colour.

Hold a subject through a crowd

People cross in front of the subject and the mask picks them up again on the other side. Long occlusions are where any mask slips, so check the whole clip rather than the frame you marked, and mark again later in the shot if it drifts.

Why rotoscoping was the slow job

A mask is only a statement about which pixels are the subject. Making that statement by hand means drawing a shape, stepping forward a few frames and correcting it, over and over, for the length of the shot. It is exact and it is slow, which is why roto has its own line on a post schedule and often its own vendor.

The shortcuts were never satisfying either. A green screen moves the work to the shoot and rules out anything you did not plan for. A keyer needs clean separation that real footage rarely has. Tracking a rectangle holds position but not shape, so the moment the subject turns, the box is wrong.

Where AI rotoscoping holds up, and where it does not

It holds up where the subject is a coherent thing with an edge, and it is unbothered by a busy background, which is where keying falls apart. Every clip on this page was tracked from a single marked frame.

It is harder where the edge itself is ambiguous. Fast motion smears a boundary across pixels belonging to both the subject and what is behind it. A subject that walks fully behind something and returns is worth marking again on a later frame. A subject that barely reads against its background is the real limit: in our tests a dancer in a translucent skirt against a near-black studio was lost after the first frame, while a yellow coat in grey fog held across every frame.

Morphic runs SAM 3.1 for this, the current version of Meta Segment Anything. For the step-by-step version, read how to track a subject with SAM 3.1.

All on Morphic

From raw clip to isolated subject

Simple pricing

Get started for free today, with the option to upgrade or cancel anytime.

Basic

$19/ month
billed as $228 per year

2400 monthly credits

1 user only

All models

Workflows

Standard

$24/ month
billed as $288 per year

3625 monthly credits

1 user only

All models

Workflows

Pro

$45/ month
billed as $540 per year

6350 shared monthly credits

1 user

+ up to 4 more at extra cost

All models

Workflows

Max

$170/ month
billed as $2040 per year

24650 shared monthly credits

1 user

+ up to 9 more at extra cost

All models

Workflows

Enterprise

For higher limits

Custom

pricing and billing terms

High-volume credits
Custom seat limits
All models
Workflows
Pricing Gradient

Free

For playing around

$0

forever free

Up to 20 credits
1 user only
Limited models
Workflows

FAQs

What is AI rotoscoping?
Rotoscoping is isolating a subject from its background so you can treat it on its own. Done by hand it means redrawing a shape around the subject every few frames, which is why it has always been the slow job in post. AI rotoscoping replaces the drawing: you point at the subject once and the model finds it and follows it through the clip, writing a mask for every frame.
How do I rotoscope a video on Morphic?
Put the clip in Compose, add a Track subject mask, and mark the subject on the frame where it reads most clearly. Click it, drag a box around it, or type a short description such as the cyclist. Pick a direction and the mask lands on the clip, ready to grade, blur, invert or combine with other masks.
Do I need a green screen?
No. That is the point of the approach. The model reads the picture and finds the subject in whatever was behind it, so footage shot in a cluttered room or on a street works the same as a studio plate. There is no chroma key to set up and no backdrop to hang.
Does the mask hold when the subject moves?
It is written per frame rather than drawn once, so it follows the subject as it walks, turns, is partly hidden and comes back. Motion blur, heavy occlusion and a subject that leaves and re-enters the frame are where any mask softens, so check the whole clip rather than only the frame you marked.
Can I rotoscope more than one subject in a clip?
Yes. Track each one separately so it gets its own mask, up to sixteen on a clip, then combine them: add two together to treat them as one, or subtract one from another to cut a hole.
What can I do with the mask once I have it?
Anything you would do with a hand-drawn one, which is the reason a mask is the output rather than a finished effect. Grade inside it, blur or pixelate what it covers, invert it to change the background, feather the edge so the cut is not hard, or stack it with a rectangle to include something the model missed.
Which model does the rotoscoping?
SAM 3.1, the current version of Meta Segment Anything. It is the same model behind Auto-segment for still images on the Canvas. See the [SAM 3.1 model page](/resources/models/sam-3-1) for what it does and the [SAM 3.1 guide](/resources/how-to/sam-3-1-guide) for a walkthrough.
Is AI rotoscoping free?
Morphic has a free tier to start. Each tracking run uses credits, and running a clip in both directions is charged as two runs. The pricing page has the current plans.

You might also like