SnapTale is an Android app that turns a child's photo into a narrated story entirely on-device. I designed and built it in five days to test whether generative AI can support curiosity and early literacy without cloud uploads, subscriptions, or text prompts.
Research → Design → Build → Closed Testing · Shipped in 5 days
I set out to answer one question: can a child aged 5–10 turn the world around them into a story — instantly, infinitely, and without ever sending a photo to a cloud?
Over a 5-day solo sprint I ran lightweight research, shipped to Play Store closed testing, and iterated with a handful of friends and their kids.
Proper research takes weeks. I had days. So I ran lightweight, mixed-method research — scoped to produce one defensible insight, not a thesis.
Quickly read up on early literacy and narrative competence in ages 5–10. Just enough theory to ground the brief — not a lit review.
Informal conversations with parents in my circle. Topics: screen-time guilt, subscription fatigue, what their kids actually use.
Heuristic walkthrough of the top photo-to-story and kids' storybook apps. Captured friction, pricing model, and privacy posture.
Watched a 5-year-old I know point at things in a garden for an hour. That one afternoon became the core insight.
"Why does the leaf curl? Is it sleepy?"
"I don't know what happens to his photos once they leave the phone."
Most ‘what is that?' moments happen at home or in the garden. A cloud dependency would kill the moment and the trust.
Monthly pricing drew instant skepticism. A one-time, zero-cost model reframes the value proposition entirely.
More options collapsed confidence. Hick's Law applies harder to this age group than to adults.
Kids didn't mind the 30–180s compute as long as something alive was on screen.
Kids spend 4–7 hours daily on screens, but they're learning less than ever.
YouTube, reels, and games are highly addictive with low educational value. Parents feel they've lost control over what their children consume — and none of the existing alternatives hold.
"When I see something cool, I want to turn it into a story I can share."
"I don't want to remove screens. I want to fix what's on them — and no safe, offline AI experience currently exists for my child."
I ran a solo Crazy-8s session, wrote four "How Might We" questions, and scored concepts against an impact × feasibility 2×2. Only one concept cleared all four HMWs.
A single-tap capture → a 4-mood picker → a visible on-device compute phase with a companion character → a narrated story saved to a visual library.
For a pre-reader, depth is the enemy of orientation. I settled on two tabs (Capture, Library) with a single linear flow under each — no hidden menus, no modals that trap.
Max depth: 3 · Average taps to first story: 3
A linear, progressive-disclosure flow. I rejected tabbed navigation mid-flow — it broke the story arc mental model for a 5-year-old.
Flows tell you where the child taps. A journey map tells you where the child is emotionally — and surfaces the risk zones a tap count can't see.
The dip at ‘Waits' is the highest-risk moment in the whole flow. Every design decision on the Processing screen is an answer to that dip — not to a technical problem.
I sketched lo-fi wireframes first, then iterated in greyscale before touching colour. The progression below shows the Capture screen's three design stages.
A single dominant circle. No toggles, no filters, no gallery button.
Contrast and hierarchy validated before any colour lands.
Yellow tab-bar added only after contrast and hierarchy were proven.
Tap targets sized against the 99th-percentile thumb reach for ages 5–10. Spacing scale follows a 4/8/16/24/32/48 step.
The shutter is the only interactive element on screen. Sized for the 99th-percentile thumb reach of ages 5–10, with no competing controls.
Pre-readers chunk into groups of 3–5. Four options kept task success high. Each option uses an icon-first, word-second format so pre-readers can choose without reading.
On-device inference takes 30–180s. Instead of hiding latency, I narrate it — invoking the Labor Illusion (Buell & Norton, 2011): visible effort increases perceived value.
The story screen serves emerging readers first — TTS narration with word-level highlighting, pause / replay per sentence, and a reading-pace slider for parents.
The library converts SnapTale from a novelty into a habit. Stories are thumbnail-first, and a soft ‘New' chip introduces a light variable-reward cue without gamification pressure.
With 5 days on the clock, I skipped formal lab usability sessions. Instead I ran closed beta on Play Store's internal track and observed live sessions with kids I know.
Quantitative benchmarks will be reported after public launch.
Running the full inference pipeline on-device isn't just engineering — it's the spine of the value proposition for parents.
Accessibility wasn't a checklist at the end — it was a design constraint that shaped the flow, the type scale, and the motion budget.
Body text ≥ 4.5:1. Story body tested at 7:1 for emerging readers.
Rounded sans, generous line-height, comfortable letter-spacing on story screens.
Minimum 48dp tap targets; shutter at 88dp; no drag gestures required.
Every story is readable and narratable via TTS with sync'd highlighting.
Companion animation respects prefers-reduced-motion.
A non-reader can complete the whole flow without typing a single character.
The app is in Play Store closed testing. Public launch metrics are not published yet, so I am tracking the signals below without presenting placeholder numbers as proof.
Public launch metrics will be published after the first cohort. Until then, the case study shows what is being measured rather than inventing proof.
"My son pointed to a mushroom on a walk and asked for ‘a story like yesterday.' That sentence alone is why this matters."
"I like that I don't have to decide whether to trust it. There's nothing to trust — nothing leaves the phone."
"He sat through the wait because the pink guy was waving. We're in trouble."
The 5-day sprint validated the core loop. These are the directions I want to explore next — each stays true to the privacy-first, zero-cost spine.
The child suggests a word; the model weaves it in — extending narrative agency.
For kids under 5 who point and speak before they tap, enter via microphone instead of shutter.
Weekly, private, on-device summary of vocabulary themes a child explored.
Export a month of stories as a printable PDF — offline, nothing sent.
The privacy-first architecture as the brief's non-negotiable. It clarified every downstream decision.
I'd run the observation session before the parent chats. The single afternoon with a 5-year-old was more useful than all my reading combined.
Waiting became a feature once it was narrated. Kids treated the compute phase like a pre-show.
A co-authoring mode where the child contributes a word and the story adapts — extending narrative agency.