Stitch Location Photos Into a Panorama for a Factual-Series Pitch
Stitch Location Photos Into a Panorama for a Factual-Series Pitch
A panorama earns its place in a factual-series pitch when it shows a location more clearly than any single frame: the shape of the room, the way one space opens into the next, where the windows sit relative to the door. It stops earning its place the moment it quietly invents something — a doorway that has been filled in, a person standing in two places at once, a room that looks sealed when it is not.
The work is three decisions taken in order. First, whether your photographs can be stitched at all. Second, how to align them and then, as a separate decision, how to display them — which projection. Third, what to do wherever the sources disagree or simply run out. Everything else — the blending, the crop, the caption — follows from those three.
This article uses Hugin to talk through the operations because its public documentation separates them clearly. The tutorial page "Stitching photos from different lenses" carries a version banner reading 0.8.0, dated July 2009, and links overlapping images through lens parameters and control points before any output projection is chosen. The manual's Mask tab page, in its "Masks" and "Masks and Control Points" sections, distinguishes blending masks from existing control points and notes what happens when an excluded region has no other source covering it. Both describe a historical build, not your installed one. Menu names move between releases. Take the concepts and locate them yourself; do not treat the labels below as a click path.
Check whether the photographs describe one compatible view
Before you open any software, make an inventory. For each photograph: where the camera stood, which way it pointed, what lens and focal length were used, what it overlaps with, and what moved while you were shooting.
Two shots taken from nearly the same spot, with the camera rotating, behave very differently from two shots taken a metre apart. A rotation sweep is the friendly case. A step sideways between frames is not, because it moves near objects against far ones. That shift is parallax, and it is geometry, not a setting you got wrong. Alignment can reconcile a rotation sweep across a moderate field of view. It cannot reconcile a lateral move for the whole frame at once, because near and far features genuinely occupy different relative positions in each shot.
Alongside the inventory, name the fixed landmarks — the features you can identify in two neighbouring frames and rely on to be in the same place: a window frame, a radiator bracket, the corner of a door lintel, the line where wall meets floor. Then name what moved. A person is never a landmark.
Here is the case I'll work through for the rest of the article. It is constructed to make the decisions visible. I have not stitched these frames, and no images accompany them.
You photographed one room — a former schoolroom, roughly rectangular, doorway on the left wall, windows opposite. One camera, one fixed focal length. No tripod: you stood near the middle and turned left to right, camera at eye level. Four frames.
- Frame 1: the left wall and the doorway, with your colleague standing in the doorway.
- Frame 2: the same left wall further along, plus the floor's tile seam. Your colleague is gone.
- Frame 3: the far wall and its two windows. Your colleague is now standing in front of the lower half of the nearer window.
- Frame 4: the second window and the wall beside it. Your colleague is gone again.
Your exposure drifted brighter across frames 3 and 4, because the camera metered toward the windows. And the far-right corner of the room — where the window wall meets the right wall — was never photographed at all.
The frames overlap in a chain: 1 into 2 along the left wall, 2 into 3 across the wall corner, 3 into 4 across the windows. That chain is compatible. Your landmarks are the window frame, the radiator beneath it, the lintel corner, and the tile seam. Your moving subject is your colleague, and your gap is the far-right corner.
Now apply the rejection test. If you had shot the left half of this room from one end and the right half from the other, five metres apart, no alignment would put the doorway in the same place twice. The honest answer there is to re-shoot, or to use the single frames instead of a composite. A stitch is only available to you when the inputs describe one compatible view. Sometimes the correct first output of this process is the decision not to build a panorama.
Align the room before trying to hide the join
Load the selected frames and check the lens information. Lens parameters describe the lens: its focal length, its distortion. With one camera and one fixed focal length, all four frames share one lens entry, which is the straightforward case. The multi-lens tutorial exists for the harder one, where two lenses or two focal lengths need separate entries and the frames still have to meet. Read the focal length from the files rather than guessing it, and if you cannot establish it, say so in the caption later rather than letting the number imply more than you know.
Control points are matched positions used to align overlapping photographs: you mark the same physical feature in two frames — the lid of the radiator in frame 3, the same lid in frame 4 — and the software uses those pairs to work out how each image maps into a shared space. Put them on fixed architecture. Do not put them on your colleague. Do not put them on the window glass, where what you are matching is a reflection of something in the room; and not on blinds that shifted, or a chair that got pushed.
Do this, and check the alignment, before choosing a projection. Alignment and projection are separate decisions, and the tutorial's order is deliberate: get the geometry to agree, then choose how to display it. If you pick the projection first, you will be looking at a beautiful display of something that is still wrong, and it will be much harder to see what.
In the constructed room, the first solve is close but not clean. The far wall tightens up nicely, the radiator lines up, and the near doorway edge does not: the jamb sits slightly further left against the wall behind it in frame 1 than in frame 2. Watch the shape of that failure. Far landmarks hold; a near feature slips. That pattern is what you'd expect from parallax — from turning around your own body rather than around the lens's entrance pupil — but it is not yet a diagnosis. Two cheaper possibilities come first.
The first is a bad control point. Suppose inspection of the point list turns up matches sitting on your colleague's shoulder in frame 1, matched to plain wall in frame 2, and a pair on the window glass. The shoulder points are wrong in principle: your colleague moved, so nothing about her position can inform how the room fits together. Worse, because the doorway is right there beside her, a point on her shoulder is pulling on exactly the part of the frame you are worried about. Remove those points, remove the reflection pair, and re-solve. In this case the left wall improves. The jamb still slips.
That leftover is the second possibility, and it is not a point problem. It is the viewpoint. The doorway jamb is close to the camera; the wall behind it is not. Two shots taken from slightly different positions disagree about where a near object sits against a distant one, and that disagreement is a fact about the capture, not an error in your work. You can add more points on the jamb and the solver will try to satisfy them, but it can only do so by dragging something else out of place — usually the far wall you just fixed. Do not spend the correction there.
Two things are worth saying plainly here. A wide, dramatic projection is a display choice, not a measurement. It stretches angles and distances by design; it is not a survey and should never be handed to anyone as a floor plan or a set drawing. And the choice of projection comes after alignment and before you start masking for exactly the reason above: it is the last decision of the alignment-and-display setup, not the thing that rescues the geometry.
Separate alignment errors from blending choices
Once the alignment is as good as the capture allows, masks decide which of your overlapping observations supplies each region of the output.
A mask is a region you draw to include or exclude an image from the blend. In the constructed room, two of them matter.
The first handles your colleague at the window. Frame 3 shows her in front of the lower half of the nearer window. If frame 4 covers that same window, exclude frame 3 from that region and let frame 4 supply the glass and the sill. But check the coverage before you rely on it: the two frames overlap across the windows, but the window she blocks is the near one, and its position in frame 4 is not something to assume. If frame 4 does not reach it, masking frame 3 does not tidy the picture — it opens a hole, and the hole is the truthful outcome. The manual page's section on uncovered regions exists precisely because a mask can remove a source without supplying a replacement.
The second mask decides a question you should ask out loud, because it is about truth rather than tidiness: where should your colleague appear in the final image? The composite is not a single instant, so it cannot show her in two positions and pretend otherwise. Pick one position she actually occupied and remove the other. In this room, keeping her at the doorway means the doorway region must be supplied by frame 1 — so you exclude the other frames from that area. Frames 2, 3 and 4 are thereby kept away from the doorway, and frame 1 supplies it.
Now the point that catches people. Masking changes which pixels reach the blend. It does not delete control points that live in the same area. If a point on your colleague's shoulder is still in the list, then masking her out of the blend leaves that point in the alignment arithmetic, still dragging the room. The manual's "Masks and Control Points" section keeps these as two separate operations for that reason. Every mask decision has a matching question about the point list, and you have to go back and answer it.
There is also a consequence people forget to look at. Choosing frame 1 for the doorway also chooses frame 1's exposure there, and frame 1 is one of your darker frames. The exposure boundary in the composite moves with the mask. That is not a mistake; it is the price of the choice, and it is something to inspect rather than assume.
What masking cannot do is repair the doorway jamb. Frame 1 and frame 2 each record a slightly different relationship between the jamb and the wall behind it. Masking lets you choose between those two recordings. It cannot produce a third one that was never captured. So the honest options are narrow: run the seam where the disagreement reads least — often along a real architectural edge, since a vertical line already in the room is forgiving — and inspect whether the jamb's thickness visibly changes across it; accept a small, visible step; crop the doorway out of the frame, which changes what the image claims; or re-shoot from a single position. In this room the jamb is not load-bearing for the pitch, so the seam is worth trying first, and worth checking before you believe it.
The far-right corner needs the same discipline. No mask can install it. Either re-shoot that corner, leave it as a visible empty area if the projection allows, or crop so the image stops before it. The crop is legitimate. What is not legitimate is a crop that gets described as a whole room.
Inspect the exported panorama against the originals
Open the export at a size where a joiner would read it — full screen — and again at the size the pitch deck will actually show it, and again zoomed into each seam.
Follow the floor's tile seam across the full width; it is a straight line in a real room, so any kink belongs to you. Check the doorway jambs for a step or a change in thickness. Check the ceiling line the same way. Look for your colleague in both places. Check for a soft vertical gradient or a hard edge where the exposure changes between frames 3 and 4 — that is a blending artefact, not an alignment error, and it wants exposure correction or a relocated seam, not more control points. Hunt for the empty region left by the uncaptured corner.
Then put the export beside the originals and confirm every spatial claim you intend to make. If the panorama appears to show a doorway that appears in no source photograph, something has gone wrong and you need to find out what before anyone else does. Compare at both large and small sizes: a defect that is invisible at thumbnail scale is a real defect in a deck that gets zoomed, and a subtle exaggeration may read fine at full size and look absurd when it is small and bright on a screen.
One more check, and it is not about the image itself. Confirm that the output has not been quietly described as an instantaneous exposure, a complete room, or a survey. Some of that work belongs in the caption.
End with the view beside its sources, and caption the limits
The package worth delivering to a pitch is not the panorama on its own. It is the exported view, the source frames that produced it, the diagnosis of what was corrected and why, and a caption that states the composite's limits in plain language: which part of the room is shown, that the people appear in one recorded position each rather than all at one moment, and where the coverage stops. "The far-right corner is outside the photographed field" is a sentence that costs nothing and prevents a much more expensive misunderstanding later. If this view later becomes a navigable tour of the location, that field-of-view statement has to travel with it. A panorama does not get more certain by being made clickable.
Success here looks modest and specific: a clearer spatial view of the room, with every correction traceable back to a source frame. Failure looks much better — a smooth, confident image that has quietly filled in the corner nobody photographed, or left your colleague standing twice.
And do not call a seam invisible until you have looked at the actual output. That is the whole habit this work is teaching: the join, the projection and the missing parts are all visible to you before they are visible to anyone else, and they stay visible only as long as you keep looking.
Frequently asked questions
When does a location panorama earn its place in a factual-series pitch?
When it shows the location more clearly than any single frame: the shape of the room, how one space opens into the next, where windows sit relative to the door. It stops earning its place if it invents something, such as a filled-in doorway, a person in two places, or a room that looks sealed when it is not.
What makes a set of photographs unsuitable for stitching?
If they do not describe one compatible view. A rotation sweep can be aligned across a moderate field of view, but a lateral move between frames produces parallax that cannot be reconciled for the whole frame at once. Shooting left and right halves from metres apart is the rejection test; the honest answer may be to re-shoot or use single frames.
In what order should alignment, projection, and masking be handled?
Align first using lens information and control points on fixed architecture, then check the alignment. Choose the output projection after alignment and before masking. Projection is a display choice, not a measurement, and it cannot rescue geometry that is already wrong.
What do masks change, and what can they not repair?
Masks choose which overlapping image supplies each region of the blend, such as excluding the frame where a colleague blocks a window if another frame covers it. Masks do not delete control points in the same area, so a point on a moving person still affects alignment. They also cannot repair a parallax disagreement in a doorway jamb or install a corner that was never photographed.
What should accompany an exported panorama in a pitch?
The exported view, the source frames that produced it, a diagnosis of what was corrected and why, and a caption stating the composite's limits: which part of the room is shown, that people appear in one recorded position each rather than all at one moment, and where coverage stops. Do not call a seam invisible until the actual output has been inspected at full size, deck size, and zoomed into the seams.