A four-corner pin and a mesh warp both make an image land where you want it. They are answering different questions, and reaching for the second when the first would do is how an hour disappears making a flat wall look worse.
What each one actually is
Corner pin is a perspective transform. You move four corners; the software solves for the single projective transformation that maps the source rectangle onto that quadrilateral. Every straight line in the source stays straight. There are exactly eight numbers involved.
Mesh warp is a per-region deformation. The image is divided into a grid, and you push individual grid points. Each cell is transformed independently, so straight lines can bend. There are as many degrees of freedom as there are grid points.
That difference — whether straight lines stay straight — is the whole basis of the choice.
The decision, in one question
Is the surface flat?
If yes, the only error is perspective, and perspective is exactly what a corner pin corrects. Use it. It will be geometrically perfect, and no mesh will improve on it.
If no, the surface has shape a projective transform cannot express, and you need a mesh. There is no amount of corner-pin fiddling that will fit an image to a curve.
Everything else is a special case of that question.
When corner pin is right
- A flat wall or screen, projector off-axis. The textbook case.
- A flat surface at an angle — a wall you are hitting from the side, a floor projection from a rig position.
- A rectangular flat object you are mapping onto — a set piece, a truck side, a banner.
- Any situation where you want the correction to be predictable and repeatable. Eight numbers can be written down and re-entered. A mesh cannot be reconstructed by hand.
When you genuinely need a mesh
- Curved surfaces. A cyc, a curved LED wall, a cylinder, a column. Nothing else will do.
- A wall that bows. Very common with hired flats and soft goods, and invisible until you project a grid on it. A 40mm bow across six metres is enough to see.
- Irregular objects. Architecture, sculpture, anything with facets you are treating as one surface.
- Edge blending between projectors. Two projectors onto one surface need local adjustment in the overlap region to align pixel-for-pixel. Corner pin gets you close; the last few millimetres in the blend need a mesh.
- Lens distortion. Wide-angle and short-throw lenses barrel slightly. On a large flat surface this shows as bowed edges that a corner pin cannot straighten.
What each one costs you
Both are resampling operations, and both cost sharpness. Where they differ is how the loss is distributed.
Corner pin softens progressively toward whichever part of the image is being stretched most. Hit a wall from a steep angle and the far side is visibly softer than the near side, because fewer source pixels are covering more surface. Nothing in software fixes that — the information is not there. The fix is physical: move the projector, or use lens shift.
Mesh warp softens per cell, and can introduce artefacts corner pin cannot: waviness where adjacent cells have been pushed inconsistently, and visible discontinuities at cell boundaries if the grid is coarse. A mesh with too many points is worse than one with few, because every point is another chance to introduce a local error.
Which leads to the most useful working rule on this subject.
Get the edges right and leave the middle alone
Audiences read the boundary of a projection against the physical object. A misaligned edge — light spilling past the flat, a gap between the image and the frame — is instantly visible to everyone in the room. A 5mm geometry error in the middle of a flat field is invisible to everybody, including you, once content is playing.
So spend your effort on the perimeter, and resist the temptation to push interior mesh points to make a test grid perfectly regular. That pursuit is the single most common way people make a map worse: chasing centre-field precision introduces waviness across the whole image that is more visible than the error it corrected.
Corollary: use the coarsest mesh that does the job. A 4 × 4 grid that fits a gentle curve is better than a 16 × 16 grid that fits it slightly better and ripples.
Do the physical work first
Both corrections cost resolution, so the goal is to need as little of either as possible.
- Get the projector square to the surface. Every degree off-axis is pixels you are throwing away.
- Use lens shift. It moves the image optically — no resampling, no loss. Spend all of it before touching software.
- Use the projector’s own geometry correction next, if it is any good. Warping in the projector keeps your canvas clean and keeps the correction with the projector rather than with the show file.
- Then correct in software, for what remains.
A projector rigged square needs a correction of nearly nothing, and looks sharper than one rigged casually and fixed digitally. This is the least glamorous and highest-value part of the process.
Test with a grid, not with content
Content hides geometry errors. That sounds like a benefit and means you discover them during the show, when something with a hard edge finally plays.
Align against a test pattern with:
- A hard outer border, so you can see the boundary against the physical edge.
- A regular grid, so local distortion is visible.
- Diagonals, which reveal bends a rectangular grid can hide.
- Some fine detail — one-pixel lines — to show you where softness lives.
Then check the result from the audience’s position, not the projector’s. From the projector, a map is correct by definition. It is the audience’s angle that reveals whether the surface was really flat.
In practice, on a normal job
Most real work uses both, in sequence: a corner pin to set the overall placement, then a light mesh for the parts of the surface that are not behaving. SpectraMap treats them that way — a placement transform per output region, with mesh adjustment available on top for surfaces that need it, rather than making you choose one model at the start.
The order that saves time: place with corners, mask the boundary, then mesh only where the grid tells you the surface is not flat. If you never need the mesh, that is a well-rigged projector, not a missed opportunity. The full order of operations is in projection mapping on a Mac, from an empty room.
Originally published on the SpectraRig blog.