Nook

• 

  Create page

• 

  Upload a photo

• 

  Select art styles

• 

  Save stickers to the album

• 

  Album page

• 

  The photo grid in the album

• 

  Sticker's detail page

• 

  Sticker's print page

• 

  Store page

• 

  Choose the sticker you want to print on the tote bag

• 

  Add the item to the shopping cart

• 

  Checkout page

• 

  User profile page

Inspiration

Nook started from a simple idea: anyone should be able to build their own creative merchandise brand.

We noticed a recurring gap. Many people have strong aesthetic taste and creative ideas, but lack design skills. Tools like Photoshop create a steep learning curve. Even if someone manages to design something, manufacturing is another barrier — factories require high minimum order quantities, inventory becomes a financial risk, and the workflow (design → sampling → supplier → shipping → storefront) is overwhelming.

We wanted to compress this entire chain into a single action: take a photo → own a product.

What it does

Nook is an AI-powered frontend application that turns everyday photos into production-ready stickers and customizable merchandise.

Core flow: Upload or take a photo → choose an art style → generate stickers → save to album → customize products → order.

• The app combines:

• AI sticker generation

• A digital sticker album

• On-demand merchandise customization

The AI identifies the main subject, removes irrelevant background elements, and applies styles such as pixel art or watercolor while preserving identity and composition. The generated result is not just decorative — it is structurally suitable for printing and die-cut production.

Users only provide creative inspiration. Nook handles design processing and production infrastructure, enabling zero-inventory, on-demand creation.

How we built it

Frontend

• React + TypeScript + Vite

• Tailwind CSS (scrapbook / journaling visual style)

• lucide-react icons

• AI & Image Processing

• Google Gemini image reasoning (subject understanding & print preparation)

Main modules

• Create — upload, background removal, style selection, batch generation

• Gallery — album management, export PDF, print preprocessing

• Store — product customization, cart, checkout

• Me — personal center & stats

Generation pipeline

1. User uploads a photo

2. AI performs subject extraction and completion

3. Multiple styles generate in parallel

4. Local canvas removes white background and trims borders

5. Stickers stored as DataURL/Blob and archived in album

Challenges we ran into

1. Reliable subject extraction

AI image models often include noise, background objects, or even add new elements. We needed the AI to:

• keep only the intended subject
• complete cropped parts
• never add text or unrelated objects We solved this through carefully engineered prompts that force strict generation rules and bias the model toward semantic focus.

1. Transparent background for printing

This was the hardest technical problem. AI models are inherently unstable at generating true transparent PNGs. Outputs were either:

• white backgrounds
• fake transparency
• edge artifacts Our solution: Generate clean white-background stickers first → remove background locally using canvas pixel processing. This gave us consistent, print-ready transparency.

1. Print consistency across products

Different items (stickers, postcards, tote bags) require different margins and composition. We created product-specific prompts so the AI composes images differently depending on the final medium.

Accomplishments that we're proud of

1. Turned casual phone photos into manufacturable sticker assets

2. Achieved consistent die-cut ready outputs

3. Built a full pipeline: creation → album → product → order

4. Produced stylized images that preserve subject identity and framing

5. Created a zero-inventory merchandise workflow for non-designers

Most importantly, users can now create something commercially usable without knowing any design software.

What we learned

We learned that building with AI is not about replacing traditional software — it’s about understanding the behavior and limitations of generative models.

Key realizations:

1. AI is excellent at style transfer with semantic awareness

2. AI is unreliable at deterministic tasks (like exact transparency)

3. Prompt engineering is effectively a form of programming

4. The best results come from hybrid systems: AI generation + deterministic local processing

Instead of forcing AI to be perfect, we designed a pipeline that lets AI do what it is good at and compensates for what it is bad at.

What's next for Nook

Our next steps focus on turning Nook from a generator into a full creator platform:

• Move storage to cloud (replace local DataURL)

• Add stronger editing tools (crop, layout, text composition)

• Complete user accounts and order management functions

• Introduce realistic mockup rendering

Our long-term goal: Nook becomes the infrastructure layer for independent creative brands — where ideas, not skills or supply chains, determine who can create.

Built With

• react
• typescript