The photobook layout problem nobody talks about

Why we couldn't just grid it

When we started work on automated photobook generation for Occasion+ users, the instinct was simple: take the photos, arrange them in a grid, print. Done. Except it isn't done. Not really.

I spent a morning looking at a handful of professional photobooks from weddings and birthday parties. The best ones weren't grids. They had rhythm. A tall portrait image would sit next to two smaller landscape shots. A full bleed photo would anchor a spread. Sometimes a single image dominated a page. Other times, five smaller moments created a gallery wall effect.

This is something a human designer internalises over years. A machine doesn't know that three square photos in a row can feel monotonous. It doesn't understand that a landscape shot after four portraits gives the eye a rest. We needed to teach it.

The data that actually matters

We started by measuring. Every photo that went into a Poolr album got analysed for orientation (portrait, landscape, square), subject matter (people, scenery, food, group), brightness, and composition. We looked at face count. We tracked which photos users marked as their favourites using the reveal mode.

Then we did something less technical: we asked. We sent surveys to wedding couples and event hosts who'd ordered photobooks. 'Which pages did you flip back to? Which arrangements felt off?' The answers weren't about grid size or image density. They were about surprise and balance. 'I loved that the page turned and suddenly there was one huge photo.' Or, 'The bit where all the candid moments were clustered together made me smile.'

One bride told us that her designer had deliberately placed a slightly blurry photo of her husband laughing next to a perfectly sharp moment of them cutting the cake. The blur made the sharpness matter more. We started thinking about contrast, not just aesthetics.

The algorithm learns through exclusion

Here's what surprised us: the layout engine spends most of its effort saying no. No, don't put four landscape photos in a row. No, don't use that blurry shot at the top of a page where it anchors the spread. No, don't cluster all the group photos together.

We fed it patterns from hundreds of photobooks and asked it to recognise what worked. But more useful than showing it good examples was showing it bad ones. A layout that put five similar photos consecutively. A spread that had no visual weight at the bottom. A sequence that never let the reader's eye settle on a single dominant image.

The code became a list of constraints that evolved over weeks. For a 40-photo photobook, there are millions of possible arrangements. Constraints narrow the field. Once we'd built the guardrails, the remaining choices all felt intentional.

One constraint surprised us: always include at least one photo that isn't of the main event. A shot of the venue entrance. A detail of the flowers. A moment of someone's shoes. Not because it's technically beautiful, but because it breaks up the sameness and reminds the reader that the experience was bigger than the obvious focal point.

The thing we didn't measure

By month two, we had a working system. We printed a test batch. Showed them to users. The feedback was honest: 'It's fine. It's... competent.' Fine. That's the word that kills you in software.

One afternoon, I was in a coffee shop watching someone flip through their photobook. They got to a page with three photos of the same moment from slightly different angles, and they laughed. A proper laugh. Not because the layout was clever, but because the three photos together told a story that no single image could. The sequence mattered.

We went back and rebuilt the scoring system to reward narrative sequence as much as visual balance. If the algorithm recognised that three consecutive uploads showed the same moment evolving, it would try to place them together, sometimes breaking its own composition rules to do it. Because the story mattered more than the grid.

That change made the biggest difference. Suddenly hosts were telling us things like, 'This feels like the day actually happened.' Which is what a photobook should be. Not a gallery. A memory.

The limits we keep

We could make this more complex. Add more rules. Let hosts tweak the algorithm. But we haven't, and we won't. There's value in constraint. When someone orders a photobook from their Occasion+ event, they get one good layout. Not ten options. Not the ability to micromanage. Just a book that arrived looking like someone who understood their event had spent time on it.

That's the decision underneath everything: we're solving for the host who doesn't want to think about layout. They've already got 200 full-resolution photos in their Poolr album. They just want them to mean something when printed.

The photobook feature sits inside Occasion+ because it's part of a bigger value proposition. Along with the live photo wall for during the event, the audio guestbook for capturing voices, the face recognition for finding yourself in the highlights. These are the tools that turn a bunch of uploads into something that feels personal.