Building an event photo slideshow that runs itself

The photographer is working the room. Guests are watching the screen to see whether they made it in. And nobody is free to sit at a laptop dragging files into a folder, because the person who would do that is running the rest of the show.

That gap — between “photos exist” and “photos are on the screen, looking good, without supervision” — is the whole job.

What “runs itself” has to mean

Four things, and if any of them fails the screen becomes a liability rather than an asset:

1. New photos appear without anybody touching the machine.
2. Bad photos do not. Test shots, lens caps, blinks, the ceiling, the shot of the photographer’s shoes while they change lenses.
3. Faces are visible. A centre crop of a group photo cuts heads off, and a group photo is most of the night’s output.
4. The screen never goes blank, including in the first ten minutes when there are three photos, and including if the photographer takes a break.

Most slideshow tools handle the first one and treat the other three as the operator’s problem. In an unattended rig, the operator is nobody.

Getting photos off the camera

Three routes, in descending order of reliability:

Tethered by cable. Camera to Mac over USB, images landing in a folder as they are shot. Utterly reliable, and it puts the photographer on a lead. Fine for a step-and-repeat or a photo booth, useless for a roaming photographer.

Camera Wi-Fi to the Mac. Most modern bodies will transfer to a computer over their own Wi-Fi or through the manufacturer’s utility. Workable, and the reliability depends on the body and on the RF environment — a ballroom full of phones is a hostile place for a camera’s low-power radio. Test it in the actual room, with the actual crowd, before relying on it.

A wireless transmitter or a Wi-Fi SD card. Purpose-built for this, generally the best of the wireless options, and an extra device to charge and configure.

Whatever the route, the outcome is the same: a folder that grows. That is the interface between the photography and the display, and keeping it that simple is what makes the rest robust. A hot folder does not care how the files got there.

The rules a hot folder needs

Watching a folder sounds trivial and has three real traps.

Partial files. A 40MB RAW takes time to write. Pick it up at 12MB and you get a corrupt image or a grey half-frame on the wall. The watcher must wait for the write to finish — either by watching for size to stabilise, or by only accepting files the transfer tool has finished renaming.

RAW plus JPEG. Most photographers shoot both. The folder gets two files per shot, and you want the JPEG. Filter by extension, and prefer JPEG when both exist for the same base name.

Order. Filesystem order is not shooting order. Sort by EXIF capture time, not by filename or modification date — filenames wrap at 9999 and modification dates reflect transfer, not capture.

Culling without a human

This is where an unattended slideshow either looks curated or looks like a memory card dump.

What can be decided automatically, in rough order of value:

• Sharpness. A simple focus measure catches motion blur and missed focus, and those are the majority of obviously-bad frames.
• Exposure. Frames that are almost entirely black or almost entirely blown are test shots, accidents, or the inside of a bag.
• Face count. On-device face detection is fast and accurate enough. Zero faces at a party is usually a room shot or a mistake; it is not necessarily bad, but it should not dominate.
• Eyes closed. Harder, and worth attempting — a blink in a two-person portrait is the shot people notice on a big screen.
• Near-duplicates. Photographers shoot three of everything. Showing all three makes the loop feel padded. Keep the sharpest of a burst.

What cannot be decided automatically: whether a photo is good. Which is fine — the goal is not curation, it is removing the frames that are visibly wrong. And there should always be a manual override, because at some point someone will need to pull one specific photo off the screen immediately, and that has to take one action.

Cropping to faces, not to centres

An event screen is usually a fixed aspect ratio, and photos arrive in portrait and landscape. Something has to reconcile them, and the default answers are both poor: letterbox everything and the screen is mostly black; centre-crop everything and you cut the top of people’s heads off, because photographers frame with headroom.

The better answer is to detect faces and let them drive the crop — keep every face in frame, weight the crop toward the faces rather than the geometric centre, and fall back to a rule-of-thirds bias when there are no faces. On a portrait photo going onto a 16:9 screen, a soft blurred fill of the same image behind is far less ugly than black bars, and it keeps the subject at a sensible size.

This is the single change that most separates a slideshow that looks designed from one that looks automatic. It is also exactly the kind of thing that is tedious for a human to do to 400 photos and trivial for the machine to do to each one as it arrives.

Music, timing and the thing that makes it feel intentional

A slideshow advancing on a fixed timer reads as a screensaver. Advancing on the beat reads as a piece of content.

It does not need to be clever. Detect the tempo of the bed music and advance on a bar boundary rather than every 6.0 seconds. Transitions that resolve on the beat rather than across it. The effect is disproportionate to the effort — people cannot say what is different, they just watch it for longer.

Two practical constraints: keep the music quiet enough to sit under room noise and speech, and be ready to mute it entirely when the formalities start. A slideshow with music playing over a speech is a mistake nobody forgives.

The screen must never be empty

The failure everyone sees is a blank or frozen screen, and it happens at two predictable moments: the first ten minutes, when there are four photos, and any point where the photo supply stops.

Both are handled the same way — have something else to show:

• Event branding, a couple’s names, a schedule card, sponsor slides.
• A pre-loaded set of photos supplied in advance. At a wedding, childhood photos; at a conference, last year’s event.
• A loop of what has already been shown, rather than stopping at the newest photo.

The rule: with fewer than some minimum number of new photos, weave branded content between them so the loop never feels thin. And never stop — a slideshow that has reached the end of its photos and is sitting on the last frame looks broken, even though nothing has failed.

Rig notes

The screen is usually not the only thing on the wall, which makes this a routing question as much as a slideshow question. If the slideshow shares a wall with presentation content, or needs to appear on a foyer screen and a stage screen with different framing, it is cleaner to treat it as a source than as an output: publish it, and let the mapping layer decide where it lands and at what geometry. SpectraMap takes it as a canvas input; Syphon or NDI depending on whether the destination is the same Mac.

Physical checks that matter for an unattended machine:

• Sleep and screen saver off, both display and system. The most common cause of a black wall at 9pm.
• Automatic updates off. An update prompt on a screen the whole room is watching is a bad evening.
• Disk headroom checked. A night of RAW plus JPEG is tens of gigabytes.
• Notifications off, or Do Not Disturb on. Whatever is on that Mac’s screen may be on the wall.

Those and the rest of the show-day sweep are in the pre-show checklist.

Where this is going for us

This is the shape SpectraPhoto is being built to: watch a folder, cull what is obviously bad, crop to faces, time to music, keep branded content in the rotation so the screen never empties, and publish the result as a source the rest of the rig can place. It is the app in the suite furthest from finished, and it is the one whose problem is most specific — every part of the list above came from watching someone do it manually at an event and thinking that should not be a person’s job.