You know the photo. Most of us have one.

Now ask yourself: do you know what she was thinking? What she found funny? What she thought about the life she lived?

To preserve a grandparent’s voice — not just their image — you need to capture three things photos can’t: the stories they told more than once, the opinions they held firmly (and sometimes wrongly), and the questions they were never directly asked. The most practical way to do this is to ask specific, story-unlocking questions in ordinary moments, record the answers, and find a format that holds the verbal and visual record together. A personalised book for grandparents built from real photos and real answers can do exactly that.

Why Do Photos Fail to Capture a Grandparent’s Personality?

A photo says: this person existed, they were here, they looked like this. A photo is proof of presence.

A story says: this person thought, felt, moved through the world, made choices, got things wrong, loved certain things for reasons that didn’t hold up under examination. A story is evidence of a person.

The difference matters because what people grieve, when someone they love dies, isn’t usually the image. It’s the conversation they can no longer have. It’s realising they can’t ask the question they should have asked ten years ago. It’s not knowing what their grandmother would have said about the thing that just happened in their life.

Photos are wonderful. But they’re a record of presence, not of personality. And personality is what we’re actually mourning.

What Makes Up a Grandparent’s Voice?

A grandparent’s voice is made up of things that don’t photograph.

The stories they told more than once. Every family has them — the one about the time they missed a train and it changed everything, or the summer they worked somewhere far from home and it was the hardest they’d ever been and also somehow the happiest. Stories that got told so many times that everyone knew the ending, but nobody interrupted. Because the telling was the point.

The opinions they held firmly and wrong. The confident but incorrect belief about cars, or politics, or how to cook a particular dish. The thing they were absolutely sure about, and which was, by any objective measure, mistaken. That wrongness was also part of who they were. We tend to sand it off in memory.

The things they didn’t say directly. What they actually felt about their own life. The choices they made and whether they’d make the same ones again. The quiet pride, or the quiet regret. This one is hard to capture because it requires asking, and we often don’t ask until it’s nearly too late.

The specific phrases. The expressions that were uniquely theirs. The ones that are now appearing in you, to your mild surprise, in the middle of a sentence.

None of these are in photos. Most of them aren’t written down anywhere either.

How Do You Actually Preserve a Grandparent’s Voice?

If you have a grandparent who’s still alive and you want to preserve something of them beyond photographs, the practical answer is simple: ask them things, and record the answers.

Not in a formal “oral history session” that feels ceremonial and a bit strange. In the ordinary moments. In the car. At dinner. When something comes up that gives them an opening.

We put together a list of 20 questions to ask your grandparents that have worked well in practice — not the abstract big ones (“what’s your philosophy of life?”) but the specific questions that unlock real stories. “What’s a job you had that I’d be surprised by?” “What did you think you’d be doing at my age?” “What were you wrong about for a long time?” Those questions work because they ask for a specific memory, not a reflection on a lifetime. They’re easier to answer.

Photos also work as prompts — not in the “what is this a photo of?” way, but in the “this photo makes me think of…” way. Show a grandparent a photo from their era and let them talk. The verbal record that didn’t fit in the frame comes out sideways.

The goal doesn’t have to be a comprehensive archive. Even three or four stories, properly recorded, is enormously more than most families end up with.

What Format Holds Both the Story and the Image Together?

The reason Memolio exists is that we wanted a format that holds both the visual and the verbal. Not just a photo album, and not just a memoir. A family memory book that captures who your grandparent actually was, told in their own words, illustrated in a way that’s specific to their face and their life.

Here’s how it works: you answer questions about your grandparent (or they answer them, with you helping). You upload a handful of photos. GPT writes a story structured around the real events and people you described. An AI vision model learns the specific faces of the people in the book. Seedream illustrates 24 pages — their childhood, the places they lived, the people they loved — with their actual face on every page. The result is printed as a hardcover book and shipped to you.

It’s not a template. Every book is different because every person is different.

We’re not open for orders yet, but if this feels like something you’d want for your own family, join the waitlist. You’ll hear about the launch first.

A Note on Timing

One of the things people say most often when this topic comes up is some version of: “I should do something about this. My grandmother is 81. I should ask her more questions.”

That instinct is correct, and it doesn’t require waiting until you have a plan. Ask one question this week. Not the biggest, most important question. Just one that might unlock a story. See what comes back.

The more we capture of a person while they’re here, the more of them we get to keep.

Frequently Asked Questions

Why is it important to preserve a grandparent’s voice, not just their photos?

Photos capture how a person looked — proof of their presence. But what people typically grieve after losing a grandparent is the conversation they can no longer have: the stories, the opinions, the specific way that person saw the world. A voice — the verbal record of who someone actually was — disappears unless it’s actively captured.

What’s the best way to get grandparents to share their stories?

Ask specific, story-unlocking questions rather than abstract ones. “What’s a job you had that would surprise me?” works better than “tell me about your life.” Ordinary moments — car journeys, Sunday meals — tend to produce better conversations than formal interview sessions. A photo from their era is often the best prompt of all.

What kinds of stories are worth capturing from a grandparent?

Stories they’ve told more than once (those are load-bearing memories), opinions they held firmly and perhaps incorrectly, things they regret or are quietly proud of, and the specific phrases or expressions that are uniquely theirs. These are the things that disappear first, and they’re rarely in any photo.

What is a personalised book for grandparents?

A personalised book for grandparents is a custom illustrated book built from real family photos and real stories about a specific grandparent’s life. Unlike template books (where names are swapped into a generic story), a personalised grandparent book is written and illustrated from scratch around that person’s actual memories, relationships, and history. Memolio makes these books using AI to write the story and illustrate each page with the grandparent’s actual face.

How long does it take to make a personalised book for grandparents?

With Memolio, the intake process takes about 20 minutes via WhatsApp or web form. The AI generates the full illustrated book in 30 minutes to 2 hours. You then review every page online, request free edits, and approve before anything is printed. Print and shipping takes approximately 7–14 business days within the EU and UK — around 2 weeks total from approval to delivery.

Am Ende hatte ich einen Zwölf-Panel-Comic. Kurz danach saß ich an Memolio.

Diese Woche habe ich erfahren, dass er einen Preis gewonnen hat. Danke an Tirol 2050 fürs Auswählen, und danke an Gloryfy für die (angeblich) unzerstörbare Sonnenbrille, die mit dem Gewinn kam. Als notorisch chaotische und unordentliche Person bin ich fest entschlossen, sie auf die Probe zu stellen.

Ich möchte über das reden, was zwischen „ich setze mich hin und mache ein einzelnes Bild“ und „am Ende sitze ich vor einer Sequenz“ passiert ist, weil ich glaube, dass es ein Muster ist, das einen Namen verdient.

Der ursprüngliche Prompt war schon ziemlich voll. Ein älterer Mann in Tracht. Kinder zu seinen Füßen. Eine retrofuturistische Farbpalette der 1950er. Solarpaneele auf Tiroler Bauernhöfen. Ich habe alles eingegeben und ein vollkommen kompetentes Bild zurückbekommen. Es war okay. Es war auch flach.

Was fehlte, war keine Qualität. Tracht und Solarpaneele rendert das Modell problemlos. Was fehlte, war Bedeutung. Das Bild sagte nichts. Es hatte keinen Anfang und kein Ende. Es wurde keine Frage gestellt und keine Antwort gegeben. Es war einfach ein Tableau.

Also habe ich ein zweites Panel hinzugefügt. Dann ein drittes. Dann habe ich weitergemacht. Am Ende waren es zwölf. Elektrobusse und Straßenbahnen, die durch grüne Stadtstraßen rollen. Seilbahnen, die vom Stadtzentrum bis zu den Berggipfeln hinauffahren. Bergdörfer, die wieder lebendig sind, weil Energie fast gratis ist. Kühe, die im Schatten von Solarpaneelen ruhen. Strom, der fast gratis fließt, wie Wasser aus einer Bergquelle. Wasser, das im Sommer in Stauseen gepumpt wird, damit Familien im Winter mit ihren Liebsten kuscheln können. Elf Panels geduldigen Aufbaus. Dann das zwölfte: Ein kleines Kind sieht zum alten Mann hoch und fragt: „Aber warum habt ihr das gemacht?“ Und der alte Mann antwortet: „Warum? Warum nicht?“

Dieser letzte Wortwechsel ist der ganze Punkt des Comics. Es gibt kein einziges Bild auf der Welt, das ihn enthalten kann. Die Frage trifft nur, weil elf Panels die Welt schon gemalt haben, über die die Frage gestellt wird. Die Antwort trifft nur, weil dir der alte Mann schon ans Herz gewachsen ist. Man braucht den Aufbau, die Pause, dann die Pointe.

Genau hier hatte ich eine leise Erkenntnis, die die letzten sechs Monate meines Lebens geprägt hat.

Einzelne Prompts sind ein One-Frame-Medium. Sie können schön sein, eindrucksvoll, manchmal verblüffend. Aber sie können strukturell keinen Bogen tragen. Das Modell weiß nicht, was davor war, und nicht, was danach kommt. Der Betrachter muss die ganze redaktionelle Arbeit im Kopf machen, und die meisten machen das nicht.

Erzählung ist die Lösung. Eine Handvoll Bilder und eine Handvoll Worte in Folge können, was keine einzelne Generierung kann. Sie können eine Frage über einen Takt halten und dann beantworten. Sie können ein Gesicht über mehrere Seiten altern lassen. Sie können einen Großvater erst Kind sein lassen, bevor er Großvater von irgendjemandem ist.

Nach dem Wettbewerb habe ich angefangen zu skizzieren, was später Memolio wurde: ein Workflow, der das Leben eines Großelternteils in ein illustriertes Buch verwandelt. Das Produkt steht auf derselben Einsicht wie dieser Zwölf-Panel-Comic. Es geht nicht darum, schöne einzelne Bilder zu erzeugen, auch wenn es das nebenbei tut. Es geht darum, sie zu einer Geschichte zu verbinden, die für eine ganz bestimmte Familie etwas bedeutet.

Tirol 2050 war ein Wettbewerb über erneuerbare Energie. Die Lektion, die ich mitgenommen habe, war eine andere und für das, was ich baue, nützlicher: KI wird in dem Moment interessant, in dem man aufhört, sie um ein Bild zu bitten, und anfängt, sie um eine Geschichte zu bitten. Der Output ist nicht mehr etwas, das man anschaut. Es ist etwas, das man liest.

Diese Verschiebung, vom Anschauen zum Lesen, ist das ganze Spiel.

I ended up with a twelve-panel comic, and not long after I was building Memolio.

This week I found out it won a prize. Thank you to Tirol 2050 for picking it, and to Gloryfy for the indestructible sunglasses that came with the win. As a deeply messy and unorganised person, I am intent on testing them to their limits!

I want to talk about what happened between sitting down with a single image in mind and ending up with a sequence, because I think it’s a pattern worth naming.

The original prompt was already loaded. An old man in traditional Austrian ‘Tracht’. Children at his feet. A 1950s retrofuturist palette. Solar panels glinting off alpine farmhouses. I fed it all in and got back a perfectly competent image. It was fine. It was also flat.

What was missing wasn’t quality. The model can render Tracht and solar panels just fine. What was missing was meaning. The image didn’t say anything. It didn’t have a beginning or an end. There was no question being asked and no answer being given. It was a tableau.

So I added a second panel. Then a third. Then I kept going. By the time I stopped I had twelve. Electric trams and buses threading green city streets. Cable cars climbing from town centres to mountain peaks. Mountain villages alive again because energy is nearly free. Cows resting in the shade of solar canopies. Near-free electricity moving like water from a mountain spring. Summer water pumped up to reservoirs so families could keep warm with the people they loved in winter. Eleven panels of patient build. Then the twelfth: a small child looking up at the old man and asking, “But why did you do it?” And the old man answering, “Why? Why not?”

That last exchange is the entire point of the comic. There is no single image in the world that contains it. The question only lands because eleven panels have already painted the world the question is being asked about. The answer only lands because you’ve already started caring about the old man. You need the build, the pause, and then the punch.

That was where I had a quiet realisation that has shaped the last six months of my life.

Single prompts are a one-frame medium. They can be beautiful, evocative, sometimes startling. But they are structurally unable to carry an arc. The model does not know what came before and does not know what comes after. The viewer has to do all the editorial work in their head, and most viewers won’t.

Narrative is the unlock. A handful of pictures and a handful of words in sequence can do what no single generation can. They can hold a question over a beat, then answer it. They can age a face across pages. They can let a grandfather be a child first, before he is anyone’s grandfather.

After the competition I started sketching what would become Memolio: a workflow that turns a grandparent’s life into an illustrated book. The product is built on the same insight as that twelve-panel comic. It isn’t really in the business of generating beautiful single images, although it does that as a side effect. It’s in the business of stringing them together into a story that means something to one specific family.

The Tirol 2050 brief was about renewable energy. The lesson I took away was different and, for the thing I’m building, more useful. AI gets interesting the moment you stop asking it for an image and start asking it for a story. The output is no longer something you look at. It’s something you read.

That shift, from looking to reading, is the entire game.

Safe to touch: most of it. Change a prompt, adjust a parameter, add a new branch. The workflow does what you'd expect. If something breaks, it usually breaks loudly.

Not safe to touch: a handful of specific Code nodes that I've come to call trouble nodes. These don't break loudly. They break silently, in a way that produces no error and no warning, and that you only discover days later when a tester reports that something is subtly wrong with their book.

Learning where these nodes live, why they behave the way they do, and how to build a working system around them has been one of the more useful engineering lessons of building Memolio, a personalised illustrated book for grandparents.

What Makes a Node a Trouble Node

The pattern is specific. A trouble node is a Code node that rebuilds its output by explicitly listing every field it wants to forward downstream, rather than using a spread or destructuring. Something like:

return {

book_id: input.book_id,

grandparent_name: input.grandparent_name,

birth_year: input.birth_year,

// ... 15 more lines

};

This exists for legitimate reasons. You don't want every piece of upstream data blindly propagating through your entire pipeline. At certain boundaries, you want to be deliberate about what flows forward.

The problem is that "deliberate" becomes "frozen." Every time you add a new field anywhere upstream of a trouble node, that field hits the explicit list, finds its name isn't on it, and disappears. Silently. The node doesn't throw an error. The next node down doesn't know to ask for the missing field. The failure is invisible until something downstream tries to use it.

The worst part is the pattern of discovery. You add a new intake question in Typeform. You wire it through the mapper. You test the mapper output and the field is there. You fire a test run and everything looks fine. Two weeks later a tester points out that a certain piece of their book is wrong, and you trace it back and find that the field died at a trouble node four hops back. The silence is the feature that makes it dangerous.

The Nodes That Keep Biting

In Memolio's workflow, the mapper is the most consistent offender. The mapper is the node that takes raw Typeform responses (or WhatsApp intake data) and transforms them into structured fields the rest of the pipeline can use. It's a complex node: it handles two intake channels, two languages, dozens of question variants, and months of accumulated edge cases.

Whenever I change a question in Typeform or WhatsApp, two things have to change: the question itself, and the mapper. If the question changes and the mapper doesn't, the field comes in with a new wording, the mapper's pattern match fails to recognise it, and it emits null downstream. No error. The field just isn't there.

This has happened several times. The mapper is a trouble node not because it hand-picks output fields in the same way as the others, but because it has an implicit field list baked into its getField() pattern matchers. Change the question wording and you've implicitly removed that field from the list.

The other consistent offenders are Prepare Book Recipe Data in WF1 (which writes the questionnaire to Supabase), and Parse Sanitized Data and Parse Story Pages (which re-emit per-page objects after LLM processing). These are all explicit hand-pick nodes. Add a field anywhere in the intake and there's a high chance it reaches one of these and stops.

The System I Built to Handle It

The first thing I did was name them. Every trouble node now has a prominent comment at the top of its code:

// TROUBLE NODE - HAND-PICK FIELD ENUMERATION

// This node explicitly lists every field it forwards.

// If you add a field upstream, add it here too or it will be silently dropped.

This sounds trivial, but it's not. When a node has that header, it becomes searchable. I can grep for TROUBLE NODE across every workflow file and immediately find all the places that need updating when I add something new. The header turns a hidden architectural constraint into a visible one.

The second thing I did was create a Claude skill that enforces the update pattern. When I'm making a change to any intake question — in Typeform EN, Typeform DE, or the WhatsApp chat engine — the skill prompts me to:

1. Update the question in the intake surface

2. Update the mapper's pattern matchers for both language variants, keeping old variants as fallbacks (historical intakes still need to parse)

3. Grep all workflows for TROUBLE NODE and check each node in the field's path

4. Add the field to every trouble node it flows through

5. Add the field to the consumer that actually uses it

6. Run an end-to-end smoke test and verify the field arrives at the consumer

The skill doesn't do the work for me. It's a checklist with context: it knows which nodes are trouble nodes, it knows the mapper has both EN and DE variants, and it knows that adding a field without updating the mapper is how we've lost data in the past.

The thing about working with AI on a complex workflow is that the AI doesn't know your architecture the way you do. I can ask Claude to add a field to the pipeline, and Claude will do it correctly at the points it can see. But it won't automatically know that Prepare Book Recipe Data needs to be updated, because that's not obvious from the code — it's institutional knowledge about which nodes are dangerous. The skill is how I've encoded that knowledge into the collaboration.

The Meta-Lesson

This is really about the gap between "works in the happy path" and "works when things change."

Any automation workflow has nodes that are fine to touch and nodes that require a ritual. The mistake is not documenting which is which, because the cost of the mistake is invisible for a long time and then suddenly very visible when a tester notices something wrong.

What I've landed on is a simple principle: if a node has an implicit or explicit field list that doesn't automatically inherit from its input, it needs to be named, documented, and covered by a checklist. The effort of maintaining the list is trivial. The effort of finding a silent data-drop bug weeks after it was introduced is not.

Building with AI tools accelerates a lot of things. The thing it doesn't automatically accelerate is the accumulation of structural knowledge about your own system. That part is still yours to do.

If you're building workflows in n8n, or anything with explicit data transformation nodes, I'd be curious whether you've hit a similar pattern and how you've handled it. The trouble node convention is working well, but I suspect there are better approaches I haven't thought of yet.

Follow the build on Substack. And if you have a grandparent whose stories deserve to be preserved, Memolio is getting close.

Memolio builds personalised illustrated books for grandparents, crafted from real memories.

My mother-in-law looked at her illustrated book and her reaction was immediate. Not “oh it’s lovely” and not “I have some notes.” Just: “I look frumpy.”

I found it funny, honestly. It wasn’t the kind of feedback I was expecting from a user test. But then my wife looked over her shoulder and agreed. And at that point I stopped finding it funny and started paying attention, because two people who know this woman well had the same reaction to how the AI had drawn her.

I went back and looked at the code.

The problem was hiding in plain sight

Every grandparent in every Memolio book — regardless of their actual age, regardless of anything the user had told us about them — was being described to the image model as “an elderly woman” or “an elderly man” in “age-appropriate casual clothing.”

Those two phrases together. That’s basically a prompt recipe for someone hunched in a cardigan.

The “how did I not see this” feeling hit pretty hard. I’d been so focused on getting the faces right, the illustration style consistent, the story prompt producing the right scenes, that I’d never stopped to think about what the model was actually being told about how this person dressed. It had never been an explicit decision. It was just a default that crept in, and I’d never questioned it.

The thing is, I had an image of “older people” in my head that came from my own grandmother. Her generation. Her clothes. And I’d just... baked that into the product without realising it.

A generational shift I’d completely missed

Here’s the thing about someone who is 65 or 70 right now: they grew up in the 60s and 70s. For a lot of people that age, fashion wasn’t just something they wore. It was a core part of their identity. They had a look. They had opinions about it. Some of them still do.

The “age-appropriate casual clothing” default had no way of knowing that. It was treating every grandparent as a generic category of person rather than as an individual who has spent decades figuring out how they want to present themselves to the world.

The fix was straightforward once I saw it: ask the user. What’s this person’s clothing style? We added a simple preference input — smart and polished, casual and relaxed, sporty, artistic, that kind of thing. Now a woman in her late 50s who describes herself as “smart and polished” gets illustrated in tailored trousers and crisp shirts instead of shapeless knitwear. Four files, about thirty minutes of work. The kind of change that probably matters more to customer satisfaction than anything else I’ve shipped this month.

But the real lesson is bigger than the fix

What I keep coming back to is how the model ended up with those assumptions in the first place. It didn’t make up “elderly in age-appropriate casual clothing” from nowhere. It learned from images. And images, across decades of media and stock photography, have a very specific idea of what an older person looks like. Dignified. Soft. Comfortable. Sensible shoes.

AI reflects the assumptions baked into the culture that produced the training data. That’s obvious when you say it out loud. It’s much less obvious when you’re in the middle of building something and just trying to get a feature to work.

This is why talking to actual users matters so much. Not surveys, not analytics, not me reviewing test books on my laptop. My mother-in-law looking at a picture of herself and saying “I look frumpy” is information I could not have generated any other way. My blind spot about what “older” looks like is a human blind spot, shaped by my own experience. The model has the same blind spot, at scale, shaped by decades of the same cultural assumptions.

The only way to find those gaps is to put the thing in front of real people and let them tell you what’s wrong.

The harder problem is still ahead

The fix I shipped is good but not complete. Asking people to describe their clothing style works when they have a clear sense of their own aesthetic. A lot of people don’t, or they describe themselves in ways that don’t map cleanly onto what the image model thinks “smart casual” or “artistic” looks like. Most people don’t neatly fit into a box.

The ideal version of this is probably more granular: specific items, specific eras, specific references. But that’s a longer form to fill in and a harder prompt to write. For now, having any preference at all is a massive improvement over a universal cardigans-for-everyone default.

Small change. Meaningful impact. One mother-in-law who might now look like herself in the book her family is making for her.

If you’re building something where AI has to represent real people, I’d genuinely like to know how you’re handling this. It feels like an unsolved problem across a lot of products. Drop a reply or find me on the blog.

And if you have a grandparent whose story deserves to be told, join the waitlist for early access when Memolio opens up properly: https://blog.memolio.io/subscribe

Memolio creates personalised illustrated books for grandparents from real memories and photos. Not yet publicly purchasable. Join the waitlist: https://blog.memolio.io/subscribe

Eine Seite aus einem Memolio-Buch — generiert mit Seedream 4.5.

Ich habe fast ein Jahr lang KI-Bildgenerierungsmodelle getestet, auf der Suche nach einem, das schöne illustrierte Seiten für personalisierte Bücher für Großeltern erzeugen kann. Das Ergebnis: Keines davon ist perfekt. Aber eines kommt am nächsten dran.

Das ist Seedream 4.5 von ByteDance. Hier ist, wie ich dort gelandet bin.

Midjourney: toll für Künstler, unbrauchbar für Produkte

Das erste Modell, das ich ernsthaft getestet habe, war Midjourney Version 7. Die Bilder sind wirklich schön. Aber ein großer Teil der besten Ergebnisse hat mit den ursprünglichen Prompts kaum etwas zu tun. Das ist kein Zufall. Midjourney ist darauf ausgelegt, dass ein Mensch mit gutem Auge die besten Bilder aus einem Stapel herausfiltert, den Rest wegwirft und von vorne anfängt. Was die Community “Prompt-Treue” nennt, ist meistens einfach Glück, das wie Absicht aussieht.

Für persönliche künstlerische Arbeit ist das völlig in Ordnung. Für ein Produkt, das eine bestimmte Szene aus dem Leben einer bestimmten Person generieren und dabei zuverlässig richtig liegen soll, funktioniert das nicht. Eine Pipeline, die auf Zufallstreffer angewiesen ist, lässt sich nicht betreiben.

Dazu kommt das API-Problem. Stand 2026 gibt es noch immer keine offizielle öffentliche API von Midjourney. Drittanbieter-Wrapper existieren, aber die laufen über Browser-Automatisierung und verstoßen gegen die Nutzungsbedingungen. Das Konto kann jederzeit gesperrt werden. Midjourney hat bewusst entschieden, künstlerische Qualität über Automatisierbarkeit zu stellen. Nachvollziehbar. Aber damit schied es vom ersten Tag aus.

Nano Banana (Google)

Googles Nano-Banana-Modelle (Nano Banana 2 und Nano Banana Pro) basieren auf Geminis Bildgenerierungsfähigkeiten.

Die Prompt-Treue ist wirklich gut. Wer etwas Konkretes beschreibt, bekommt es. Für einen Anwendungsfall, bei dem der Prompt viel tragen muss, also Haarfarbe, Kleidung, Zeitepoche, Setting, mehrere Figuren, ist das wichtig.

Zwei Dinge haben mich trotzdem gestoppt.

Erstens die Ästhetik. Das Herbst-Campus-Bild weiter unten zeigt das Problem: schöne Komposition, angenehme Farben, aber eine unangenehme Gleichmäßigkeit. Jedes Element gleich stark ausgearbeitet, kein Papier, das durchscheint, keine weichen Kanten, an denen Farbe ausläuft. Es sieht genau so aus, wie die meisten Menschen sich “KI-Bilder” vorstellen. Für ein Buch, das jemand ins Regal stellt und aufhebt, ist das ein Problem.

Nano Banana 2. Eine KI-typische Wiedergabe einer Person.

Zweitens die Kosten. Nano Banana 2 kostet 0,08 $ pro Bild, Nano Banana Pro 0,15 $. Seedream 4.5 kostet 0,04 $. Bei 24+ Illustrationen pro Buch und bis zu 50 kostenlosen Nachbesserungen während der Review-Phase summiert sich das schnell. Ich habe Nano Banana Pro kurz für die Darstellung von Seitentiteln getestet, aber bei 0,15 $ pro unzuverlässigem Versuch ist ein sauberes CSS mit gutem Webfont die bessere Wahl.

Flux

Flux Pro hat einen angenehmen Illustrationsstil und ordentliche Prompt-Treue. Aber das Bild unten zeigt das Kernproblem: gute Komposition, schöne Stoff-Textur, vollkommen generisches Gesicht. Das Modell ließ sich von den Referenzfotos inspirieren, anstatt sie wirklich zu lernen. Bei einem Produkt, bei dem das Buch wie die konkrete Person darin aussehen soll, ist das ein grundlegendes Versagen.

Flux Pro. Schöner Stil. Falsche Person.

Flux 2 ist auf andere Weise schlechter. Die Ergebnisse wirken wie KI-Einheitsbrei: zu glatt, zu sauber, alles an der richtigen Stelle, aber ohne die kleinen Unvollkommenheiten, die ein Aquarell menschlich machen. Ein Vergleich mit dem Seedream-4-Alpenbild weiter unten macht den Unterschied klar. Das eine sieht gemalt aus. Das andere sieht generiert aus.

Flux 2. Technisch korrekt. Völlig leblos.

Grok / Aurora

Die Bildqualität ist technisch in Ordnung: gute Prompt-Treue, Ästhetik besser als bei vielen Konkurrenten. Aber ich habe eine persönliche Abneigung dagegen, Elon Musks Unternehmungen zu finanzieren. Seine Plattform hat jahrelang Hetze verstärkt. Während des Bundestagswahlkampfs hat er sich aktiv eingemischt, wiederholt die AfD beworben und Alice Weidel auf X eine große Bühne verschafft. Das ist für mich keine abstrakte politische Position, sondern eine konkrete Entscheidung: Ich möchte nicht, dass Geld, das ich für API-Aufrufe ausgebe, dieses Ökosystem mitfinanziert. Andere werden das anders abwägen, und das Modell ist technisch legitim. Für mich ist es aus der Liste, unabhängig von der Bildqualität.

Seedream 4

Bei Seedream 4 habe ich zum ersten Mal etwas gefunden, das ich bei keinem anderen Modell gesehen hatte: Es sah tatsächlich aus wie ein Aquarell.

Ein Hinweis zu den folgenden Bildern: Sie zeigen alle mich, in verschiedenen Altersstufen. Der Junge bin ebenfalls ich.

Es gibt eine Wärme und Weichheit in der Darstellung, eine handgemachte Qualität, die ich anderswo nicht gefunden habe. Das Alpenbild unten zeigt das gut: Der Berghintergrund löst sich auf, anstatt hart abzuschneiden. Das Gras an den Füßen läuft in den Hang über. Für ein Buch, das eine Oma in die Hand nimmt und ihren Enkeln zeigt, zählt das mehr als technische Präzision.

Seedream 4. Wunderschöne Komposition, aber die Ähnlichkeit ist generisch.

Das Problem war die Zuverlässigkeit. Seedream 4 produzierte eine schöne Seite und dann, bei der nächsten Generierung, etwas anatomisch Falsches. Das Wasserpistolen-Bild unten ist ein gutes Beispiel: schöne Komposition, tolle Sommerenergie, warmes Nachmittagslicht. Der Junge im Vordergrund hat außerdem drei Arme.

Seedream 4. Der Junge im Vordergrund hat drei Arme.

Sobald der Prompt eine komplexere Anordnung beschrieb, also mehrere Figuren, ungewöhnliche Winkel, aktive Szenen, konnte das Modell auseinanderfallen. Und wenn die Referenzfotos 50 Jahre alt sind und aus einem echten Fotoalbum gescannt wurden, hat das Modell weniger zu lernen und die Ergebnisse werden unzuverlässiger. Das ist kein Randfall für dieses Produkt. Die meisten Menschen, über die wir Bücher machen, haben vor dem Digitalfoto-Zeitalter gelebt.

Der Wechsel zu 4.5

Ich habe den Wechsel länger hinausgezögert als nötig. Die ersten Ergebnisse von Version 4.5 haben mich nicht so gepackt wie Version 4, und ich habe das zunächst als Qualitätsverlust gedeutet. War es nicht. Die Ästhetik ist etwas anders, und ich musste mich daran gewöhnen. Was sich tatsächlich verbessert hat, ist die Prompt-Treue: Die anatomischen Fehler und Kompositionsprobleme aus Version 4 treten in 4.5 deutlich seltener auf.

Das Campus-Bild unten zeigt das in der Praxis. Die Figur ist in Bewegung, die Ähnlichkeit mit dem Referenzfoto ist besser, und die Aquarellqualität ist noch da: die Farbspritzer im Himmel, der negative Raum auf dem Weg, die Hintergrundfiguren, die angedeutet statt vollständig ausgearbeitet sind.

Seedream 4.5. Bessere Prompt-Treue, eine treffendere Ähnlichkeit, und die Aquarellqualität ist noch da.

Vollständig behoben sind die Probleme noch nicht, weshalb wir einen Review-Prozess gebaut haben, bei dem jede Seite von einem Menschen geprüft wird, bevor irgendetwas in den Druck geht. Die Leute vertrauen uns ihre Familiengeschichten an. Eine Oma, die ein Buch bekommt, auf dem sie drei Hände hat, ist nicht akzeptabel.

Eine Komplikation, die ich nicht erwartet hatte: ethnische Drift. Seedream wurde überwiegend mit asiatischen Daten trainiert, und ohne explizite Angaben neigt das Modell dazu, Gesichter mit ostasiatischen Merkmalen zu generieren. Für ein Buch über eine Oma aus dem Bayerischen Wald oder aus Hamburg ist das ein offensichtliches Missverhältnis: nicht weil etwas falsch wäre an diesen Gesichtszügen, sondern weil sie diese Menschen schlicht nicht darstellen.

Die Lösung war ein optionaler Schritt im Onboarding, bei dem Kunden ihren familiären Hintergrund beschreiben können, ergänzt durch ein Vision-Modell, das Referenzfotos analysiert und physische Merkmale für den Bildprompt extrahiert. Das funktioniert einigermaßen gut. Aber die Erfassung ethnischer Herkunftsdaten bedeutet die Verarbeitung besonderer Kategorien im Sinne der DSGVO, was ausdrückliche Einwilligung und sorgfältigere Handhabung erfordert als ein normales Datenfeld. Ein kleiner technischer Fix, der eine nicht ganz kleine Compliance-Frage aufgemacht hat.

Warum Seedream 5 nicht funktioniert hat

Ich habe es getestet. Die Ergebnisse waren enttäuschend.

Seedream 5. Jeder Pixel gefüllt. Kein Raum zum Atmen.

ByteDance hat mit 5.0 Lite einen bewussten Kompromiss gemacht: Das Modell ist auf kommerzielle Produktfotografie, Werbemotive und Poster ausgelegt. Der Fotorealismus ist besser. Aber der Vergleich mit dem Seedream-4-Alpenbild oben sagt alles: Jeder Zentimeter der Fläche ist ausgefüllt. Die Jacke hat einzeln ausgearbeitete Knöpfe. Die Stadt unten hat vollständig detaillierte Ziegeldächer. Kein negativer Raum, kein durchscheinendes Papier, nirgendwo Luft zum Atmen. Es fühlt sich an wie ein Foto, über das jemand Pinselfarbe gestrichen hat, nicht wie etwas, das wirklich gemalt wurde.

ByteDance hat angekündigt, die Ästhetikfrage beim vollständigen 5.0-Release nochmal anzugehen. Für die Lite-Version gilt: falsches Werkzeug für illustrierte Familienbücher.

Stand der Dinge

Seedream 4.5 ist meine aktuelle Antwort. Keine perfekte: Kein Modell ist das, weshalb die Bearbeitungsfunktion existiert und weshalb jede Seite vor dem Druck von einem Menschen geprüft wird.

Der Modellvergleich, der in diesem Bereich wirklich zählt, sind keine Benchmark-Werte. Es geht darum, wie das Ergebnis für eine 70-jährige Oma aussieht, die ein Buch über ihr Leben in der Hand hält. Die meisten dieser Modelle wurden nie für dieses Publikum entwickelt. Seedream 4.5 kommt dem am nächsten, was ich bisher gefunden habe.

Ich baue Memolio — personalisierte illustrierte Bücher für Großeltern, entstanden aus echten Fotos und echten Erinnerungen. Jedes andere personalisierte Buch für Großeltern setzt ihren Namen in eine erfundene Geschichte. Memolio erzählt ihre wirkliche Geschichte. Wenn du das für deine Familie möchtest, trag dich in die Warteliste ein.

A page from a Memolio book — generated with Seedream 4.5.

I’ve spent the better part of a year testing AI image generation models, trying to figure out which one can produce beautiful illustrated pages for personalised books for grandparents. The answer is that none of them do it perfectly. But one gets closest.

That’s Seedream 4.5 from ByteDance. Here’s how I got there.

Midjourney: built for artists, not products

The first model I used seriously was Midjourney, version 7. It produces genuinely beautiful images. But a lot of those images had almost nothing to do with the prompts that generated them. That’s not an accident. Midjourney is still largely built around a human with taste plucking the good outputs from a batch of generations, discarding what doesn’t work, and trying again. What the community calls “prompt adherence” is less the model following your instructions and more you getting lucky with a beautiful accident.

For personal artistic work that’s completely fine. For a product that needs to generate a specific scene from a specific person’s life and get it right, it doesn’t work. You can’t build a pipeline around hoping.

There’s also the API problem. As of 2026, Midjourney still doesn’t have an official public API. Third-party wrappers exist, but they use browser automation and violate the terms of service. Your account can be banned at any point. Midjourney has made a deliberate choice to optimise for artistic merit and manual creation rather than enabling products. I respect that. But it ruled them out on day one.

Nano Banana (Google’s Gemini image models)

Google’s Nano Banana models (Nano Banana 2 and Nano Banana Pro) are built on top of Gemini’s image generation capabilities.

The prompt adherence is good. When you describe something specific, they execute it. For a use case where the prompt is doing heavy lifting (describing hair colour, clothing, era, setting, multiple characters), that matters.

Two things stopped me from using them.

First, the aesthetic. Look at the autumn campus image below. The scene is well composed. The colours are nice. But there’s an evenness to everything: every element rendered to the same level of finish, no paper showing through, no soft edges where pigment might bleed. It looks like what most people picture when they think “AI image.” For a book people are going to keep on a shelf, that’s a problem.

Nano Banana 2. A ChatGPT rendition of a person.

Second, the cost. Nano Banana 2 is $0.08 per image, Nano Banana Pro is $0.15. Seedream 4.5 is $0.04. When you’re generating 24+ illustrations per book and offering up to 50 free regenerations during the review phase, that difference compounds fast. I tried Nano Banana Pro briefly for rendering page titles into the illustrations, but at $0.15 per attempt for something that inconsistent, you’re better off writing CSS and using a font you can actually control.

Flux

Flux Pro has a pleasant illustration style and reasonable prompt adherence. But look at the figure below: well composed, nice rendering of fabric, completely generic face. It took inspiration from the reference photos rather than learning from them. For a product where the book is supposed to look like the specific person it’s about, that’s a fundamental failure.

Flux Pro. Nice style. Wrong person.

Flux 2 is worse in a different way. The outputs look like AI slop: too smooth, too clean, every line in the right place and none of the imperfect details that make a watercolour feel handmade. Compare the three life-stage panel below to the Seedream 4 Alpine image further down. One of them looks painted. The other looks generated.

Flux 2. Technically correct. Completely soulless.

Grok / Aurora

The output is technically decent: reasonable prompt adherence and an aesthetic that works better than most. But I have a personal aversion to funding Elon Musk’s ventures. His platforms have spent years amplifying hateful content, he’s shown active enthusiasm for political destabilisation in Europe, and his general direction of travel is one I’d rather not contribute to financially. Other people will weigh that differently, and it’s a legitimate model to test. It’s off my list for reasons that have nothing to do with image quality.

Landing on Seedream 4

Seedream 4 was where I found something I hadn’t seen in any other model: it actually looked like watercolour.

A note on the images that follow: they’re all of me, across different ages. The boy is me too.

There’s a warmth and softness to the rendering that gives pages a handmade quality I haven’t found elsewhere. Look at the Alpine image below: the mountain background dissolves rather than terminates, the grass at his feet bleeds into the slope. For a book a grandparent is going to hold and pass to her grandchildren, that matters more than technical fidelity.

Seedream 4. Beautiful composition, but the likeness is generic.

The problem was reliability. Seedream 4 would produce a beautiful page and then, on the next generation, give you something anatomically wrong. The water gun fight below is a good example: lovely composition, great summer energy, the warmth of late afternoon light. The boy in the foreground also has three arms.

Seedream 4. The boy in the foreground has three arms.

Any time the prompt described a complex arrangement, multiple characters, unusual angles, active scenes, it could fall apart. And when reference photos are 50 years old, scanned from an actual photobook, the model has less to work with and the results get less reliable. That’s not an edge case for this product. Most of the people we’re making books about lived before the era of digital photography.

Moving to 4.5

I held off on the upgrade longer than I should have. My first outputs from 4.5 didn’t grab me the way version 4 had, and I initially read that as a quality drop. It wasn’t. The aesthetic is slightly different, and I needed to adjust to it. What actually changed is prompt adherence: the anatomical failures and composition collapses that plagued version 4 are materially reduced in 4.5.

The campus image below shows what that looks like in practice. He’s actually moving, the likeness is closer, and the watercolour quality is still there: the splashes in the sky, the negative space on the path, the way the background figures are suggested rather than fully rendered.

Seedream 4.5. Better prompt adherence, closer likeness, and the watercolour is still there.

Not that the problems are gone entirely, which is why we built a full review pipeline where a human checks every page before anything goes to print. People are trusting us with their family stories. A grandparent receiving a book where she’s depicted with three hands is not acceptable.

One complication I didn’t anticipate: ethnicity drift. Seedream is trained predominantly on Asian data, and without explicit guidance, the model tends to render faces with East Asian features by default. For a book about someone’s grandmother who grew up in rural Bavaria or suburban London, that creates an obvious mismatch: not because there’s anything wrong with those features, but because they don’t represent who these people actually were.

The fix was an optional intake step where customers can describe their family background, alongside a vision model that analyses reference photos and extracts physical descriptors to pass into the image prompt. It works reasonably well. But collecting ethnic background data means collecting a special category under GDPR, which requires explicit consent and more careful handling than a standard data field. A small technical fix that opened a non-trivial compliance question.

Why Seedream 5 didn’t work

I tested it. The outputs were disappointing.

Seedream 5. Every pixel filled. No room to breathe.

ByteDance made a deliberate trade-off with 5.0 Lite: it’s designed for commercial product photography, advertising layouts, poster design. The photorealism is better. But look at that image and compare it to the Seedream 4 Alpine scene above: every inch of the canvas is filled. The jacket has individually rendered buttons. The town below has fully detailed rooftiles. There’s no negative space, no paper showing through, nowhere for the eye to rest. It feels like a photo with paint brushed over the top of it rather than something actually painted.

ByteDance have said the full 5.0 release will revisit this. As of the Lite version, it’s not the right tool for illustrated family books.

Where things stand

Seedream 4.5 is my current answer. Not a perfect one: no model is, which is why the editing feature exists and why every page needs a human in the loop before it gets anywhere near a printer.

The model comparison that actually matters in this space isn’t benchmark scores. It’s how the output looks to a 70-year-old holding a book about her life. Most of these models have never been optimised for that audience. Seedream 4.5 gets closer than anything else I’ve tried.

I’m building Memolio — personalised illustrated books for grandparents, made from real photos and real memories. Every other personalised grandparent book puts her name in a made-up story. Memolio puts her actual story in a book. If that sounds like something you’d want for your family, join the waitlist.

Sie hatte erzählt, wie sie ihren Mann kennengelernt hatte. Beide waren Lehrer, sie trafen sich in der Schule, in der sie unterrichteten. Als die KI diese Szene illustrierte, zeichnete sie die Testerin als Kind. Ein Mädchen am Pult, nicht eine erwachsene Frau vor der Klasse.

Sie hatte geschrieben, dass sie Lehrerin war. Es stand wörtlich im Text. Aber irgendwo zwischen ihren Worten und dem fertigen Bild hatte die KI eine Annahme getroffen. Schule plus Begegnung gleich jung. Sie hat gemustert, nicht verstanden.

Und das ist die Sache mit KI: Sie hat nicht denselben Kontext wie ein Mensch. Wenn jemand sagt: „Wir haben uns in der Schule kennengelernt”, fragt ein Freund automatisch nach. Als Schülerin, als Lehrerin, als Eltern am Elternabend? Eine KI wählt einfach die statistisch wahrscheinlichste Lesart und macht weiter. Sie hält nicht inne. Sie weiß nicht, was sie nicht weiß.

Ich habe Monate damit verbracht, dieses Erinnerungsbuch für Großeltern so zu bauen, dass es jedes Mal perfekte Ergebnisse liefert. Bessere Anweisungen, präzisere Prompts, Validierungsschritte, Fallback-Logik. Und es ist deutlich besser geworden. Perfekt? Nein. Weil menschliche Leben schön, unordentlich und nie ganz einzuordnen sind. Eine Oma, die in den Siebzigern Bergsteigerin war. Ein Opa, der sein Leben lang hawaiianische Hemden getragen hat. Ein Paar, das sich 1968 auf einer Demo in Frankfurt kennengelernt hat. Das sind keine Ausnahmen. Das sind genau die Details, die ein Familienerinnerungsbuch wertvoll machen. Und es sind genau die Details, die eine KI am ehesten in etwas Allgemeines einebnet.

Diese Erkenntnis hat mir eine Weile Angst gemacht. Ich hatte das Einbauen einer Bearbeitungsfunktion lange vermieden. Es fühlte sich an wie ein Eingeständnis, als würde ich zugeben, dass die KI es allein nicht schafft. Aber die Enttäuschung dieser Testerin hat das Bild umgedreht. Die Bearbeitungsfunktion ist kein Rückzieher. Sie ist das Produkt, so wie es funktionieren sollte. Ein Mensch erzählt seine Geschichte, die KI macht daraus einen ersten schönen Entwurf, und dann verfeinert ihn derjenige, der die Geschichte wirklich kennt.

Entscheidend war für mich der Gedanke daran, wer ein personalisiertes Buch für Großeltern eigentlich bestellt. In Deutschland sind das meistens Töchter und Söhne, die ihrer Mutter oder ihrem Vater etwas zum 70., 75. oder 80. Geburtstag schenken wollen. Oder Enkel, die zu Weihnachten etwas Persönlicheres suchen als eine weitere Tasse. Diese Menschen sind nicht in der KI-Welt zu Hause. Sie werden keinen besseren Prompt schreiben und sich nicht in die Technik einarbeiten. Aber sie wissen haargenau, dass ihre Mutter keine Schülerin mehr war, als sie ihren späteren Mann traf. Sie wissen, welche Farbe die Küche hatte, in der sonntags gegessen wurde. Die Bearbeitungsfunktion gibt ihnen einen Weg, dieses Wissen ins Buch zu holen, ohne auch nur eine Zeile Code oder einen Prompt schreiben zu müssen.

Die breitere Lehre, und sie gilt, glaube ich, für alle, die gerade mit KI etwas bauen: Hör auf zu versuchen, die Maschine perfekt zu machen. Mach es stattdessen den Menschen einfach, das zu korrigieren, was die Maschine falsch macht. Genau dort liegt die eigentliche Magie. Nicht in einem makellosen ersten Ergebnis, sondern in der Zusammenarbeit zwischen menschlicher Erinnerung und maschineller Kreativität.

Das ist es, was ich bei Memolio baue. Ein personalisiertes Erinnerungsbuch für Großeltern, in dem echte Geschichten und KI-Illustration zusammenkommen. Es ist noch nicht kaufbar. Wer möchte, kann sich auf der Seite in die Warteliste eintragen und den Weg mitverfolgen. Diese eine Rückmeldung hat alles daran verändert, wie ich über das Produkt nachdenke.

Gartner now predicts that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. UiPath launched industry-specific agents. Every automation platform from Zapier to n8n is repositioning as an "agent platform." And some experts are predicting the first $1 million business fully run by AI agents before the year is out.

Here's the thing about the agentic AI hype: it's simultaneously true and misleading. The technology is real. AI agents can browse websites, compare options, book things, and chain together complex tasks. I use them every day to build Memolio — they write code, migrate files, debug infrastructure, and draft content.

But "agent" suggests autonomy, and autonomy suggests you can step away. In practice, you can't. Not yet.

The honest answer is that agents have blinders on. Context windows keep growing — we're at a million tokens now and still climbing — but a simple overconfidence in approach can negate that advantage entirely. An agent that doesn't think to search for up-to-date information, or that pattern-matches to a familiar solution when the situation actually calls for something different, can spin in circles indefinitely. This week's Lambda debugging episode was a case study in exactly that.

But the deeper issue isn't capability — it's comprehension. An agent can manage a process brilliantly. It cannot build a process from scratch with the same intuition a human brings, because it doesn't understand the full context of a business: the emotional core, the messy human relationships, the things that matter even though they've never been written down. Something will always be missing. The question is whether you've designed your system so a human catches what's missing before it causes a real problem.

Only about 21% of companies have successfully deployed AI workflows at enterprise scale. The other 79% are stuck on orchestration and governance — a polite way of saying they can't figure out when to let the AI run and when to pull the handbrake.

At Memolio, the entire product pipeline is a chain of AI agents — just not the kind VCs talk about at conferences. An n8n workflow takes a grandparent's answers from WhatsApp, generates a personalised story with GPT, creates custom illustrations with BytePlus Seedream, assembles a print-ready PDF, and sends it to a printer. Each step is automated. Each step is also carefully constrained.

The n8n architecture wasn't a clever design decision made upfront — it was built out of necessity. Memolio is a product about memory and emotion. When you give an agent too much freedom with that material, it goes wrong in ways that feel genuinely uncomfortable. It draws people who never existed. It illustrates scenes in ways that hurt the memory of a real person rather than honour it. It interprets a grandparent's answer in a way that shifts the emotional meaning entirely. These aren't minor bugs — they're violations of the thing the product is supposed to do.

So every node is designed to produce one specific thing in a specific way to get a predictable result every time. The structure isn't a constraint on the AI. It's the whole point.

This week's Lambda debugging episode captured the tension perfectly. My AI coding assistant had all the tools it needed: web search, code execution, AWS documentation. It was, by any reasonable definition, an "agent." And it still looped on the same wrong fix five times because it lacked the meta-cognition to say "this approach isn't working, let me try something fundamentally different."

Here's my honest take: AI agents are one of the most remarkable technological developments of my lifetime. I say that as someone who went from being full of ideas with no way to implement them, to being on the verge of having a real product that does something that genuinely matters. Without agents, I don't get here. Full stop.

But I haven't had a single agentic work session run to completion without some human intervention. Not one. And I think that's the gap the industry hasn't fully reckoned with yet.

One of the biggest things I've been working on is an edit function — the ability for a grandparent or family member to look at an illustrated page and say "no, that's not right, that's not how it happened." Building that layer — the human-in-the-loop, the edit-and-iterate function — is often harder and more important than the AI generation itself.

The most valuable AI isn't the most autonomous AI. It's the most interruptible AI. The system that runs reliably until it can't, then stops and asks for help — instead of confidently doing the wrong thing twenty times in a row.

The VCs are betting on removing humans from the loop. I'm betting on knowing exactly where in the loop humans need to stay.

What's your experience? Are AI agents living up to the promise, or are you still babysitting? I'd love to hear — especially from other indie makers and solopreneurs building real products with these tools. Drop a comment or find me at memolio.substack.com.

I knew him as the man who’d built a beautiful garden out of an empty plot in his retirement. I knew him as the person who could hold forth on the Boer War or the Second World War for hours — which, on long car journeys as a seven-year-old, I would encourage enthusiastically to the quiet despair of my brother and sister, who did not share my morbid fascination with industrial-scale conflict. I knew him as a voice on the phone that came less frequently as I grew older and the cost of flying from London to Cape Town began to feel like a luxury rather than a given.

But I didn’t know him. Not really. And standing in that sea of black clothes and sad faces, I couldn’t work out exactly when the gap had opened up.

The Gap That Grows Quietly

My mother is South African. I grew up in the UK. In childhood, distance was just a fact — we visited every few years, and in those visits I’d get to ride on his shoulders and pepper him with questions about history that he’d answer patiently and with obvious delight. But as I got older and started paying my own way, the visits became rarer. And somewhere in that drift, the questions I had for him — the ones that actually mattered — became harder to ask.

Not because he wouldn’t have answered. But because asking felt forced. It felt like imposing. He was a humble man who seemed to think his own life was too ordinary to be worth a proper conversation. The grand historical sweep was easier territory than the personal — he could talk about the Huguenot migration or Rommel’s campaign without it feeling awkward. But asking “what were you like when you were young?” without a reason or a context behind it? That requires something.

A seven-year-old asking questions out of pure curiosity is easy. An adult asking the same questions has to overcome years of accumulated social training that says nosiness is rude and other people’s feelings are fragile. The window for those conversations is surprisingly small, and once it closes, it closes quietly.

The Book on the Shelf

After the funeral, I found a family history book he had put together in his final years. It was, like him, meticulous and humble in equal measure. Chapters and chapters on the Huguenot migration, the Anglo-Boer War, genealogies of relatives long dead — all carefully researched, all beautifully written. But the parts that stopped me in my tracks were the sections where he talked about his own early life: collecting bird eggs as a boy, struggling to find a career that felt right, falling in love with my grandmother, having five children in quick succession and somehow making it work. He’d written these sections almost apologetically, always steering the narrative back toward the people around him rather than himself.

Reading it, I felt two things at once: profound gratitude that he’d written it at all, and a deep sadness that I hadn’t read it while he was still alive. The parts I’d have asked him about, the details I’d have wanted to see — his young face, the texture of the world he’d grown up in, the emotion beneath the careful, modest sentences — those were gone. That’s where the idea for Memolio came from.

Why It Has to Be a Book

There’s no shortage of ways to preserve memories digitally. Voice recordings, photo albums, Google Drive folders. The problem isn’t the technology. It’s that digital things disappear — not always dramatically, but reliably. Files get buried. Hard drives fail. Cloud storage runs out or changes its pricing model. A video of your grandfather explaining how he met your grandmother sits unseen in a folder you haven’t opened in four years.

A physical book doesn’t do that. A physical book sits on a shelf and waits. It’s there when a child is three and wants to look at the pictures. It’s there when they’re seven and can finally read the words themselves. It’s there when they’re seventeen, or twenty-five, or forty, and suddenly — for reasons they might not be able to articulate — want to understand where they came from.

At Memolio, we take a grandparent’s real memories and real photos and turn them into a personalised illustrated story — the kind that a grandparent can give to a grandchild and read together at bedtime. The illustrations bring the stories to life in a way that old photographs, however precious, never quite can: faces infused with the emotion of the moment, scenes that make a past world feel present and real. The questions we ask grandparents are designed to draw out exactly the parts they’re most likely to gloss over — the personal things, the human things, the early life things. Not because their family history isn’t interesting, but because they are interesting, and they often need permission to say so.

What I Hope Memolio Does

I hope that someday, someone stands at a grandparent’s funeral and feels not just the weight of their absence — but also the warmth of their lingering presence. That they go home and pick up a book they read with their mum or dad about who that person was before they were a grandparent. That the stories don’t disappear with the person who lived them.

That’s the bridge. Not technology. Not a clever product feature. Just stories — told with love, held in hands, passed forward.

If you have a grandparent whose stories deserve to be heard — or if you are the grandparent — we’re building this for you. Follow along at memolio.substack.com, or visit memolio.io when we launch.

The problem itself was boring: I’d deployed a small AWS Lambda to process PDFs as part of our book printing pipeline. It worked perfectly when I tested it inside the AWS console. The moment I called it from anywhere else — my own computer, our automation tool, anywhere that wasn’t the console itself — it returned 403 Forbidden. Locked out by its own configuration.

Claude’s suggestion was confident and clear: “Just re-add the permission, it’s a 2-minute fix.” I did. It didn’t work. “Try deleting the old permission first, then re-adding it.” I did. It didn’t work. “Let’s use the CLI instead of the console.” I did. It didn’t work. Every fix came with the same breezy certainty, and every failure was met with a new suggestion delivered in the same breezy certainty.

At some point I noticed something that should have been obvious from the start: we were going in circles. Claude wasn’t tracking what had already failed. It was pattern-matching each new attempt to a different prior problem, confidently prescribing solutions that I had already tried, sometimes twice. I was the memory. I was the thing keeping the debugging session coherent, and I was terrible at it because I was tired and it was 1am and I wanted to go to bed.

Eventually I got genuinely annoyed and typed something I don’t usually type at a computer:

“you really need to get better at recording what HASNT worked when debugging because we are just going round in circles”

That single sentence flipped the entire session.

What changed when I forced the step back

Claude wrote a file called DEBUG_403.md. Inside it: every attempt, every result, every hypothesis still alive. It read the actual error messages character by character instead of skimming them. It did a proper web search. And within about ninety seconds it had the real answer.

Turns out AWS quietly changed how Lambda Function URLs work in October 2025. They used to need one permission (lambda:InvokeFunctionUrl). They now silently require two (lambda:InvokeFunctionUrl AND lambda:InvokeFunction). If you only have the first one, every public request returns 403, forever. The AWS console had actually been telling me this the whole time — there was a little blue banner explicitly stating both permissions were needed. I read it nine times over the course of the debugging session. Claude generated a response to it nine times. Neither of us actually understood what it said, because we were both too busy pattern-matching to similar-looking past problems.

The fix was one CLI command.

The thing I keep relearning

I’ve been working with AI coding agents intensively for almost a year now. I build Memolio almost entirely with them — most days I don’t write code directly, I describe what I want and review what comes back. And I keep learning the same lesson in different forms: AI agents will loop on an approach forever if you let them. Sometimes the most valuable thing a human can do is force a step back.

It’s tempting to read this as “AI is unreliable, don’t trust it.” That’s not quite right. Claude wasn’t being lazy or malfunctioning. It was doing what large language models do: generating the next most plausible response given the state of the conversation. The problem was that the state of the conversation kept looking like a problem it had almost-seen-before, and “almost-seen-before” is a bad foundation for debugging something you haven’t actually seen at all.

What the model needed was a different kind of context: not more suggestions, but an enforced structure. A file on disk listing what had been tried. A requirement to read the error message literally before guessing. A rule that says “on the second failed attempt, stop guessing and start researching.” None of this is beyond the model’s capabilities — it did all of it brilliantly once asked. It just wouldn’t do it on its own.

I think this is the actual skill of working with AI agents in 2026. Not prompt engineering. Not picking the right model. Not even building clever evaluation loops. It’s knowing when to interrupt. Knowing when the confident-sounding next step is actually the fourth iteration of the same wrong step, dressed in different words. Knowing when to stop letting the machine drive and make it justify itself.

When I did that last night, Claude wrote me a memory file — literally, a permanent note to its own future sessions — called feedback_record_what_hasnt_worked.md. The rule: “On the second failed attempt at any fix, stop guessing. Create a debug log. Read error messages literally. Do proper research before suggesting more button-clicking.” It will remember this next time I start a new session. I don’t know if that’s exactly what machine learning researchers mean when they talk about AI that learns from feedback, but it felt like something.

The thing that actually shipped

Oh, right. The Lambda works now. The entire print pipeline for Memolio is automated end-to-end for the first time. When a grandparent finishes their WhatsApp interview and approves their book in the review page, everything from “interview complete” to “physical book printed and shipped” now runs without a human touching it. A webhook fires, PDFs get generated, the Lambda adds the tiny invisible printer metadata that commercial presses require, real MD5 hashes get computed, the order lands in CloudPrinter’s system, and forty-five seconds later there’s a confirmed sandbox order waiting to go.

And it happened in ninety seconds once we stopped pretending we knew the answer.

Memolio turns a grandparent’s stories into a personalised, illustrated book — captured over WhatsApp, illustrated by AI, printed and shipped in hardcover. We’re currently in closed user testing. If you want early access or just want to follow along as we build, you can reply to this email or sign up at memolio.io.

Until last week, it was called “Who Is Grandpa?”

Then I killed the name.

Not because it was bad. It worked. People understood it immediately. But it had a problem I couldn’t ignore: it only described half the people buying it.

Our actual customers are mostly women — mums, daughters, granddaughters — buying a family memory book for the whole family. “Who Is Grandpa?” left them out. And it forced us into four separate domains across two languages (English and German), which is an SEO nightmare nobody warns you about when you’re picking a product name at 2am.

So I started searching. Over a hundred domain names later — Latin portmanteaus, German compound words, every “love” and “keepsake” variant you can imagine — we landed on Memolio. Memo + folio. Memory + pages of a book. Short, gender-neutral, works in every European language. One domain instead of four.

But renaming a product does something unexpected. It forces you to sit with the question: what is this thing, really?

Strip away the name, the domain, the logo. What’s left?

A bet. A bet that the stories our grandparents carry — stories about wars survived, countries left behind, trades learned, loves lost and found — deserve better than a Facebook post that disappears in a week. Better than a conversation that happens once and then lives only in the memory of whoever was in the room.

My generation is probably the last one that will hear these stories directly. Our grandparents lived through things most of us can barely imagine, and those stories live in their heads — not written down anywhere. Every year, more of them are lost forever. Not because nobody cares, but because nobody thought to ask in time, or nobody knew how to turn the answers into something lasting.

That’s what Memolio is for. You don’t need to be a writer. You don’t need to sit down with a tape recorder and transcribe hours of conversation. You answer a few questions — through WhatsApp or a simple form — share some photos, and we use AI to turn those answers into a personalised illustrated children’s book. Watercolour illustrations of real moments. Real places. Real people.

Not a template where we swap in a name. A completely unique story, built from the life your grandparent actually lived.

The technology behind it is genuinely interesting — I’m building it with n8n workflow automation, AI image generation, and a lot of duct tape — and I’ll write about that too, in the “Build in Public” section of this newsletter. But the technology is not the point. The point is the book on the shelf. The point is a four-year-old asking to hear the story again. The point is connecting generations in a way that lasts longer than a phone call.

We’re not live yet — still in testing, still fixing bugs, still learning from real users what works and what doesn’t. But if this resonates with you, I’d genuinely love to hear about the family story you’d want to preserve. Reply to this post, or just think about it next time you’re with your grandparents.

Some stories are too good to lose.

— H