IMIP Artefact Demo - 8 May 2026

Arclight

An AI-assisted capture-to-sticker archive for everyday objects.

Artefact + Report15-minute demoiOS-first app
From a single object capture to a lightweight visual memory.
BFL-generated sticker style contact sheet
Creative Memory

A sticker is useful when the object keeps its context.

Arclight combines sticker generation, style transfer, and editable object metadata inside a personal archive.

A generated sticker becomes meaningful when the object remains editable and explainable.

Creative memory here means preserving a small everyday object as a visual marker with usable context.

Arclight supports AI generation and style transfer, but it also treats each object as a record whose metadata can be reviewed and changed.

01
AI draft

Recognition creates initial object metadata such as name and category.

02
User edit

The user can correct or add metadata after generation.

03
Memory record

The sticker keeps context through editable information, not only visual style.

Prototype Evidence

All prototype screenshots, grouped by surface

31 device screenshots from V1 and V2. Click a surface to inspect its iteration in detail.

Click a surface to inspect its iteration in detail.
V1 Homepage 01.1 V1 Homepage 01.1
V1 Homepage 01.2 V1 Homepage 01.2
V1 Homepage 02 V1 Homepage 02
V1 Homepage 03.1 V1 Homepage 03.1
V1 Homepage 03.2 V1 Homepage 03.2
V1 Homepage 04.1 V1 Homepage 04.1
V1 Homepage 04.2 V1 Homepage 04.2
V1 Homepage 05.1 V1 Homepage 05.1
V1 Homepage 05.2 V1 Homepage 05.2
V2 Homepage 01 V2 Homepage 01
V2 Homepage 02 V2 Homepage 02
V1 Library 01 V1 Library 01
V1 Library 02 V1 Library 02
V2 Capture 01 V2 Capture 01
V2 Capture 02 V2 Capture 02
V2 Capture 03 V2 Capture 03
V2 Review 01.1 V2 Review 01.1
V2 Review 01.2 V2 Review 01.2
V2 Review 01.3 V2 Review 01.3
V2 Review 02 V2 Review 02
V2 Review 03 V2 Review 03
V2 Review 04 V2 Review 04
V2 Review 05 V2 Review 05
V2 Review 06.1 V2 Review 06.1
V2 Review 06.2 V2 Review 06.2
V2 Detail 01.1 V2 Detail 01.1
V2 Detail 01.2 V2 Detail 01.2
V2 Detail 02.1 V2 Detail 02.1
V2 Detail 02.2 V2 Detail 02.2
V2 Calendar 01 V2 Calendar 01
V2 Calendar 02 V2 Calendar 02
Optional Prototype Inspection

Homepage Deep Dive

The home surface moved from explanation and camera entry into a visual sticker/object board.

The home surface moved from explanation and camera entry into a visual sticker/object board.

  • V1 explained the collection idea and pushed camera access.
  • V2 makes stickers and object labels the first visible evidence.
  • The final direction treats Home as archive surface, not only navigation.
V1 Homepage 01.1
V1 Homepage 01.2
V1 Homepage 02
V1 Homepage 03.1
V1 Homepage 03.2
V1 Homepage 04.1
V1 Homepage 04.2
V1 Homepage 05.1
V1 Homepage 05.2
V2 Homepage 01
V2 Homepage 02
Optional Prototype Inspection

Library Deep Dive

The separate Library surface was removed from the V2 structure and merged into Home.

The separate Library surface was removed from the V2 structure and merged into Home.

  • V1 explored categories and saved item browsing.
  • The final direction reduces the separate library layer.
  • Home becomes both display surface and archive entry point.
V1 Library 01
V1 Library 02
Optional Prototype Inspection

Capture Deep Dive

Capture became deliberately narrower: fewer controls, stronger real-object behaviour.

Capture became deliberately narrower: fewer controls, stronger real-object behaviour.

  • Earlier capture concepts carried extra affordances.
  • V2 makes the camera button the dominant action.
  • The simplification supports real-world object capture.
V2 Capture 01
V2 Capture 02
V2 Capture 03
Optional Prototype Inspection

Review Deep Dive

Review became the place where AI output is inspected before it becomes a memory record.

Review became the place where AI output is inspected before it becomes a memory record.

  • Cutout and recognition are visible before saving.
  • Style transfer is chosen rather than hidden.
  • The screen turns AI output into an editable decision point.
V2 Review 01.1
V2 Review 01.2
V2 Review 01.3
V2 Review 02
V2 Review 03
V2 Review 04
V2 Review 05
V2 Review 06.1
V2 Review 06.2
Optional Prototype Inspection

Detail Deep Dive

Detail supports object metadata, not just a full-size generated image.

Detail supports object metadata, not just a full-size generated image.

  • AI creates an initial name/category suggestion.
  • The user can correct metadata after generation.
  • Location, notes, and category make the sticker easier to revisit.
V2 Detail 01.1
V2 Detail 01.2
V2 Detail 02.1
V2 Detail 02.2
Optional Prototype Inspection

Calendar Deep Dive

Calendar makes saved stickers visible as moments in time.

Calendar makes saved stickers visible as moments in time.

  • Saved stickers become temporal memory markers.
  • The archive can be revisited through dates.
  • This supports the creative-memory framing.
V2 Calendar 01
V2 Calendar 02
Iteration

From camera entry to object-first archive.

The prototype moved toward fewer navigation layers and stronger object evidence.

Homepage From

Camera entry / explanation

To

Visual sticker board
The opening surface became a place to display saved stickers and object memories.

Library From

Separate category page

To

Merged into Home
Removing the extra surface reduced navigation weight and made the archive feel immediate.

Capture From

Import, flash, and options

To

Focused photo button
The simplified capture flow encourages photographing real-world objects.

Project Aims

The aim is an editable object memory, not just a generated image.

The artefact must capture, understand, restyle, edit, and preserve the object.

01Capture a real everyday object
02Isolate the object from the image
03Recognise it and create initial metadata
04Let the user edit that metadata
05Generate or transfer a chosen sticker style
06Save it as a local object memory record
System Pipeline

The artefact is a pipeline, not just an image prompt.

Each step converts a real object into an editable saved record.

01
Camera capture
The user photographs a real everyday object.
02
iOS Vision cutout
Foreground extraction runs locally in the iOS development build.
03
Gemini metadata
Recognition creates an initial name, category, and object context.
04
Review / edit
The user checks the cutout and edits the AI metadata draft.
05
Style choice
The user selects the visual direction before generation.
06
FLUX Worker
The style Worker creates the generated sticker image.
07
SQLite / FileSystem archive
Records and files persist as an editable local object memory.
Design Decisions

The interface keeps AI editable.

The design choices reduce friction while keeping the user in control of the object record.

Object-first Home

Home displays stickers and saved objects, not only an entrance to capture.

Library merged into Home

Fewer navigation layers make the archive faster to understand.

Simplified Capture

A focused camera action lowers complexity and encourages real-object photography.

Review as control point

AI creates a draft, but the user confirms metadata and chooses style.

Local editable archive

Saved items remain revisitable records, not disposable AI outputs.

Live Demo Map

The live demo proves the editable loop.

I will show the path from real object capture to a saved sticker memory.

Demo flow 1: Home 01 Home
Demo flow 2: Camera 02 Camera
Demo flow 3: Photograph 03 Photograph
Demo flow 4: Review 04 Review
Demo flow 5: Metadata 05 Metadata
Demo flow 6: Style 06 Style
Demo flow 7: Save 07 Save
Demo flow 8: Board 08 Board
Demo flow 9: Calendar 09 Calendar
Demo flow 10: Detail edit 10 Detail edit
Switching To Phone

Capture. Check. Revisit.

The slide deck pauses here while the artefact becomes the evidence.

Checkpoint 01Capture

Photograph a real object in the room.

Checkpoint 02Check

Review recognition, metadata, cutout, and style.

Checkpoint 03Revisit

Save the record, then show Home, Calendar, and Detail.

Implementation Evidence

Implemented as an iOS-first production prototype

Native processing, Worker-proxied AI services, local persistence, and automated checks support the demo.

App

Expo 54, React Native 0.81, Expo Router 6

Native

Local Expo module, iOS Vision subject cutout

AI services

Cloudflare Workers for Gemini and FLUX

Storage

Expo SQLite records and FileSystem images

Verified

npm run lint passed.

npx vitest run passed with 21 test files / 84 tests.

This is implementation evidence for the artefact prototype, not formal user research.

Limits

The main limits are pipeline and metadata conditions.

The prototype is demo-ready, with clear dependency and evaluation limits.

iOS development build

The full pipeline depends on the iOS 17 development build and native module.

Configured Workers

Recognition and style generation require valid Worker URLs and API keys.

Network availability

AI calls need network access, so a backup recording remains useful.

Metadata reliability

AI recognition can be wrong, so Review/Edit is part of the core design.

Evaluation scope

Current evidence is smoke testing and automated tests, not broad user research.

Future Work

Next: evaluate, broaden, stabilise.

The current prototype establishes the loop; the next work should make it more robust and better evidenced.

01User testing

Observe whether editable sticker memories support real creative-memory use.

02Platform

Broaden beyond the iOS-first prototype with Android support.

03Control

Improve metadata editing, style consistency, and offline fallback.

Closing Claim

Arclight turns object captures into editable sticker memories.

Capture, metadata, style, and local archive work together as the artefact.

BFL-generated sticker style contact sheet
Questions

Any questions?

01 / 14
Arclight - IMIP Demo