Masks, and the honest cost of rotoscoping

A mask is a shape that says where

Draw a shape; an effect applies inside it, or outside it. That is the whole idea, and it underlies windows in grading, blurring a face, replacing a sky, and every composite in this block.

Three properties matter on every mask:

• Feather — how soft the edge is. A hard edge always reads as cut out. Almost every mask wants some.
• Expand or contract — moving the edge in or out by a few pixels, which is how you remove a fringe of background that came with your selection.
• Invert — apply outside instead of inside.

Static, tracked, animated

A static mask sits still. Fine for a locked-off shot: a vignette, darkening a blown window, blurring a fixed sign.

A tracked mask follows something the tracker can lock onto. Fine for a rigid object moving through frame.

An animated mask has its shape keyframed over time, because the thing it covers changes shape as well as position. This is rotoscoping, and it is expensive.

Rotoscoping, honestly

Rotoscoping is drawing a mask around a moving subject, frame by frame, for the length of a shot.

A realistic rate for a person walking, done by hand with a bezier shape and keyframes every few frames: somewhere between five and thirty seconds of work per frame of footage, depending on the complexity of the edge. A five-second shot at 25 fps is 125 frames. That is one to four hours for five seconds.

This number is worth knowing because it changes decisions. Rotoscoping is not a thing you casually do to fix a shot; it is a budgeted task, and in professional work it is frequently outsourced to specialists.

What reduces the cost:

• Keyframe sparsely. Set a shape every 5 to 10 frames and let interpolation fill between, then fix where it drifts. Do not draw every frame.
• Split the subject into parts. A separate shape for the head, the torso, each arm. Each moves more simply than the whole, so each needs fewer keyframes.
• Track what you can. Attach a shape to a planar track for any part of the subject that moves rigidly.
• Use the shot for less time. Two seconds needs half the work of four.

And the thing that reduces it most: do not rotoscope if a different solution exists. A cutaway, a different take, a reframe, or simply not doing the effect.

Hair, motion blur and transparency

The three things a drawn mask cannot represent.

Hair is thousands of fine strands; a bezier shape approximates it as a smooth outline and the result reads as a helmet. The professional answer involves a separate hair matte and considerable effort.

Motion blur means the subject's edge is genuinely semi-transparent across several pixels. A hard shape has no way to express that, so a rotoscoped moving limb looks cut out.

Transparency — glass, smoke, water, a net curtain — is not a shape at all.

When the shot contains any of these, either the effect changes or the shot does.

Machine-assisted rotoscoping

Resolve has a Magic Mask that segments a person, or a part of a person, using a machine-learning model: paint a stroke on the subject and it produces a matte and tracks it. Similar tools exist elsewhere, and Segment Anything-derived models power several free implementations.

This is genuinely useful and it changes the economics for a certain class of shot. Two honest caveats. It is good at people and body parts and poor at arbitrary objects. And the matte it produces is soft and imprecise at exactly the edges that matter — hair and motion blur — so it gets you 80% of the way and the last 20% is still hand work.

On Resolve, some Magic Mask features are Studio-only depending on version; check yours before planning around it.

The common jobs

Blurring a face or a licence plate. A tracked shape with a mosaic or blur. Use a solid shape rather than a light blur if the intent is genuine anonymity, and check every frame — the track will slip.

Sky replacement. A luminance key usually beats a drawn mask, because the sky is bright and uniform. Problems are at the horizon and in anything thin against the sky — branches, wires, hair.

Removing something at the edge of frame. Often better solved by a reframe than by a mask.

A split screen. Two rectangular masks and no difficulty.

Today

Draw a mask around a moving object for one second of footage with keyframes every five frames. Time yourself. That number is the one to quote from.