CushLabs Sticker Gen
Turn AI-generated art into real WhatsApp sticker packs your friends can actually install — about five minutes from raw image to live in your chats.
The Problem
You ask an AI for a sheet of cute stickers and it happily delivers — except every sticker is welded onto a fake checkerboard background, wrapped in a white outline the exact same color as everything around it, and WhatsApp flatly refuses to accept any of it. There is no "send to WhatsApp" button anywhere. Getting from "look how adorable these are" to "they're actually in my chats" turns out to be a surprisingly deep rabbit hole — so I built the little factory that runs the whole trip for you.
The Solution
Instead of fighting to erase the background, the pipeline keeps only the artwork and draws a crisp new sticker outline around it — so checkerboards and shadows simply can't sneak through. It figures out the grid on its own and keeps each sticker's little extras (hearts, zzz's, confetti, captions) attached to the right character, then bundles everything into one tap-to-install app that WhatsApp actually trusts — and that quietly refreshes itself whenever you add a new pack. From a raw AI sheet to live-in-your-chats is about five minutes.
Key Features
- 18 packs / 216 stickers shipped from AI sheets, every sheet carved clean with zero manual masking
- Handles cartoon, dark-emblem, and photoreal art across 3×4 and 4×3 grids with grid auto-detection
- Synthesized die-cut borders via distance-transform offset curves after proving color-keying impossible
- Signed Android app (one APK, all packs) distributed via a trusted GitHub Releases CDN that beats Android download-blocking
- Live bilingual (en / es-MX) marketing site with per-pack carousels, full SEO, and ~5-minute new-pack turnaround
Results
Overview
CushLabs Sticker Gen converts AI-generated sticker sheets into a real, distributable WhatsApp sticker product. A Python pipeline segments each sheet and rebuilds clean die-cut borders; a customized build of Meta's official sample app packages every pack into one signed APK; and a bilingual Astro site at stickers.cushlabs.ai distributes it. It began as a single axolotl pack and became a repeatable factory now shipping 18 packs / 216 stickers, each new pack roughly five minutes from raw sheet to live download.
The Challenge
Source PNGs have no real alpha — transparency is a baked-in checkerboard, or a flat tint more saturated than the artwork's own whites. White die-cut borders are pixel-identical to the background, so no color threshold separates them. Satellite elements (hearts, ZZZ, confetti, multi-line caption text) are disconnected components that must follow their parent sticker. Sheets arrive in different grids and wildly different art styles with no metadata. And WhatsApp validates everything on-device, only accepts packs from an installed app, and caches them by version — there is no folder to drop files into.
The Solution
Carve, don't key. The pipeline keeps only colored/dark artwork pixels, then synthesizes the white border as a distance-transform offset curve, so background and shadows can never survive into the output. A flat-background mode classifies pixels by their ratio to an edge-sampled background color for tinted sheets.
Robust assignment. Grid dimensions are auto-detected from sticker-centroid clustering (a transposed command is silently corrected), and satellites attach agglomeratively nearest-first, so even caption-text chains anchor to the right sticker.
Real distribution. One signed APK serves all packs through WhatsApp's official handshake. The bilingual Astro site distributes it via a public GitHub Releases mirror — a trusted CDN that sidesteps Android's download-blocking of unknown hosts — and bumps image_data_version so updated packs auto-refresh in WhatsApp without users re-adding them.
Technical Highlights
scipy.ndimageconnected-component segmentation; distance-transform border synthesis- Two background models, per-component noise filtering, grid auto-detect, agglomerative satellite chaining
- Static + animated pipelines (green-screen MP4 → ≤500 KB animated WebP)
- Multi-pack
contents.jsonmerge with correct cache-version handling - Astro 6 / Tailwind 4 bilingual site: hreflang, sitemap, JSON-LD, obfuscated email, pack carousels, security headers
- Local-keystore signing with file-based password; per-cell QA overlays
Results
- For users: 18 private, ad-free sticker packs added to WhatsApp in one tap, in English or Mexican Spanish — no third-party sticker apps, no community feeds, no re-encoding.
- Technical demonstration: end-to-end ownership of a hard image-processing problem (multiple failed segmentation strategies diagnosed with pixel-level evidence before the synthesis approach), plus full product delivery — Android toolchain, signing, a trusted-CDN distribution fix for real download failures, and a deployed bilingual SEO site.
Ready to discuss a similar solution?
Let's explore how AI automation can help your business.
Schedule a Consultation