Short answer: A matte painting shot follows a fixed sequence. The plate and camera data come in, the lens distortion is measured and removed, the camera is rebuilt inside the straightened image, reference is collected, a rough blockout gets approved, the painting is created in depth layers and projected onto simple geometry from the solved camera, the distortion is reapplied, the plate's grain and optics are matched in the result, and the work is delivered as an image sequence. If a stage is skipped, the cost moves to the end of the schedule, where it is highest.
Most breakdowns of this work present the final frame separated into layers. For a producer, that does not explain what to send, when to review it, or why one shot costs more than the shot next to it. This article follows the process in order instead. At each step, keep two questions in mind: what the production must provide, and what it receives back. If you want the definition rather than the sequence, what a matte painting is covers that, and what a set extension is covers one common version of the job.
What has to arrive before anyone paints
No work begins before the shot exists as material. Usually, that involves five things.
The plate is the ungraded filmed footage, kept at the resolution the show finishes at, and it includes whatever the camera recorded outside the framing rather than being cropped tight to it. That margin proves useful in the next section.
Details for the camera and lens. A focal length alone says very little, because the effect a lens has on a frame depends on the relation between that number and the size of the sensor or film gate behind it, so the format must accompany the number. Add the camera height and angle, the focus distance and the stop, any survey or lidar from the location, and a distortion grid, defined as one frame of a printed chart shot on the same lens at the same settings. Shooting that grid is a small request on the day, and it removes an entire category of argument later.
The duration of the shot as it is cut, plus the handles, the extra frames on either side that remain available to the edit.
The colour setup consists of the working space, the conversion that takes the camera's own encoding into that space, and any grade or show look provided as a separate file, not baked into the pixels.
In addition, the approved design contains concept art, production design reference, and stills from the other shots at the same location.
You get back a breakdown and a list of questions at this stage, not an image. If a vendor asks nothing here, they are guessing, and that guess shows up late in the schedule as a revision the production pays for.
The lens first, then the camera
Real lenses make straight lines curve. Camera solvers expect a lens that does not, so distortion is measured before or during the track rather than after it. Nuke's LensDistortion node estimates this distortion either through Grid Detection or manual Line Detection; the grid exists for that purpose. Its output modes include undistort, redistort, and an STMap that stores the warp itself as an image.
That STMap is worth naming specifically. It lets the painter, the CG department and the compositor use the same exact warp instead of three near matches.
Straightening the lines recovers the camera: where it was, how it moved, what it looked through. According to Foundry, CameraTracker is a matchmoving tool that lets an artist build a virtual camera that moves the same way as your original camera. You get back a camera file, usually Alembic or FBX. Before you send it, agree on the scene scale and where the world origin sits, because a camera solved in one unit next to a survey built in another looks right and measures wrong.
The paint is affected in two ways. Removing distortion pushes the image outward, which makes the working frame larger than the delivery frame; every layer must be painted into that margin, or the redistort at the end drags black in at the corners. A locked-off camera reduces this whole stage to almost nothing, which is why a vendor asks about the camera move before the view.
Reference, then a blockout nobody is supposed to like
Collecting references means choosing a photograph whose conditions match the plate, not one that is beautiful: the sun angle, haze, weather, and hour all need to line up. A golden hour reference shot is of no help for a plate filmed under flat overcast, and using it anyway makes a frame end up looking painted.
Next there is the blockout: an intentionally crude version of the entire frame. Grey shapes are placed at the correct distances, set up in the solved camera, and rendered at the delivery format, so the composition you approve is the composition that ships. It is supposed to look unfinished.
The point of the blockout is that the expensive mistakes happen while they are still cheap. Composition, scale, the horizon line, whether the mountain appears five kilometres away or fifty. If those decisions come late, they lead to a repaint. The blockout in the shot comes back as a still and a short movie. The production supplies an answer, and a slow one here holds up all downstream work.
The painting, built in pieces from the first stroke
After the blockout receives approval, the image is built from paint, photographic material, and rendered elements in any case where a specific behavior is required.
From the start, the scene is split by depth instead of being built as one picture: sky, far distance, middle ground, near ground, foreground. Each layer is painted in full, even the parts that sit behind the layer in front of it, because a moving camera will reveal them. Much of that hidden area never appears on screen at all, and this is one reason a still and a shot are priced differently.
Two points concern the method of painting, not the subject. The projection, the redistort, and the final reformat each resample the image and none of them restore detail, so layers have to be built above delivery resolution. In most paint packages, what you see is a display image rather than the underlying scene values, so any material painted through a viewing transform must be returned to the show's working space deliberately. Careless work appears first in highlights. A finished still never reveals the depth separation in digital matte painting, and the number of layers comes down to judgement rather than a formula.
Putting the flat image back into space
Simple geometry receives the projected layers. In Nuke, the Project3D node operates exactly as its name implies: it projects an input image through a camera onto the 3D object.
With the solved camera parked on a selected frame, the image is projected from it; the shot is then viewed through that same camera as it moves. Freezing the wrong frame leaves the painting accurate for one moment of the shot and smeared across the rest. Multiple frames are projected and blended across the geometry whenever a single projection cannot cover the move.
The underlying geometry tends to be basic. The usual setup is cards and a ground plane, trimmed to the key silhouettes and placed at the distances the solve indicates are real. If a shot truly requires a modelled, lit and rendered environment instead, that is a different service on a different cost curve, and 3D environment work covers it.
From this point forward, review the shot in motion. A frame that looks stable as a still can shift against the plate once the camera moves, and you can only see that movement during playback.
Integration, which is where the shot is actually won
The image is correct now, but the shot is still incorrect because the painting is cleaner than the surrounding photography. During integration, the new material inherits every flaw the plate already has, and this occurs in sequence.
First the painted element has to be placed at the correct depth against the filmed plate. This means using holdout mattes and roto for everything it passes behind, and giving its edges the plate's own defocus and motion blur wherever they cross a foreground. A shot fails first at that boundary, and a soft edge crossing it, like foliage or hair, is a different job from a hard architectural edge.
The optical work comes next. Haze in the air thickens over distance at the same rate as the plate's own haze does. Toward the frame edges, real lenses become softer. Color fringing. Around bright sources, bloom appears. The camera actually recorded a black level rarely as dark as a painting's starting black.
After the distortion is switched back on, the frame is cropped from the oversized working format to the delivery format.
Apply grain after that step, and this order is not optional. Grain is a property of the sensor, so it already exists in the camera's distorted image; grain added before the redistort gets stretched and softened by the warp. Match it afterward, channel by channel, to the size and strength the plate actually has.
Colour underpins the entire process, so a shot is not judged on an artist's monitor. More productions now use managed setups, and the Academy Color Encoding System describes itself as an industry standard for managing color and digital files throughout the lifecycle of almost any media production, partly to remove ambiguity when work passes between multiple vendors. For a buyer, that means supplying your colour configuration and viewing transform on day one, not at delivery.
Review, and what final delivery actually contains
Reviewers review in versions, and the usual chain starts internally, then goes to the visual effects supervisor, then to the production, and then to the director, with the notes becoming smaller as it progresses. They watch it at speed on the largest screen available, through the show's viewing transform, because that is where an error in haze or grain shows up and a desk monitor hides it.
What is delivered is an image sequence, not a picture file. The return uses the plate's resolution and frame range, includes handles, stays in the show's colour space, follows the naming convention the show issued, and includes a review movie with the viewing transform baked in.
OpenEXR is a common container here; according to its documentation, pixel data are saved as 16-bit or 32-bit floating-point numbers and a file can contain any number and combination of image channels. Because of this second property, a vendor can deliver the separated layers together with the finished frame, so the show's compositors can adjust the shot later without returning to the painter. Ask for this at the bid stage. It is more difficult to arrange later.
Why does the same note cost nothing at one stage and a repaint at another?
What a note invalidates determines its price, not how it is worded.
Adding a note on the blockout costs almost nothing. No downstream parts exist yet.
After painting begins, any change to the design costs the layers it touches and all the layers behind them, since moving an object in the middle ground uncovers distant background that was painted to remain hidden.
The note on light is the expensive one. Light is not a separate layer; it is present in every stroke, in the haze, in the black levels, and in the edges, so a new sun direction means a new painting that keeps the old composition.
After you solve the camera, a note placed on the plate resets the solve, the undistort, and the projections underneath. The painting on top survives only in the places where the framing did.
If a camera move is inserted during the edit once projections have been set, a shot that previously required one projection now requires several. The image itself stayed the same. What changed was the method.
A production controls that part of the cost most directly. It depends on the order in which the answers arrive, not on the amount of work in the frame.
Who paints these
Based in Toronto, Chumak Studio creates painted environments for film and television. The work is by Anastasiia Chumak, an artist with 31 film and television credits from 2017 to 2026, including House of the Dragon, "Senior matte painter, Pixomondo", and Daredevil: Born Again, "Senior matte painter, Folks VFX". You can view the full list on the credits page.
If you cannot tell which of these stages applies to your shot, send the plate and the camera information through contact and you will receive a direct answer about whether this tool fits it, including when it does not.
