SnapSmart – Full-Page Capture + On-device AI

Chrome extension for flawless full-page screenshots with on-device Gemini Nano AI: summarize pages, rewrite text, and write short drafts. Includes PDF capture mode and works fully offline.

Comment

Inspiration

Taking a clean, full-page screenshot is still surprisingly hard—pages jump, sticky headers overlap, infinite scroll breaks captures, and PDFs inside the browser are awkward. I wanted one tool that (1) always produces a perfect stitched image and (2) lets me use Chrome’s built-in AI (Gemini Nano) locally to summarize, rewrite, and write without sending data anywhere.

What it does

SnapSmart is a Chrome extension that:

Captures pixel-perfect full-page screenshots with overlap-aware stitching and a smart Freeze that pauses animations, sticky headers, and lazy loading.

Offers Manual Stitch for infinite scroll feeds and tough pages.

Adds PDF Capture Mode for direct .pdf links.

Runs on-device AI in the popup:

Summarize visible page text (or summarize the last capture via a multimodal fallback).

Rewrite any pasted/selected text.

Write short drafts from a prompt.

Privacy-first: AI runs entirely in Chrome, no server calls.

How we built it

Stitching pipeline: step-scroll the viewport → capture → crop top/bottom overlaps → draw to a canvas → final trim to exact content height.

Freeze mode (content script): temporarily disables CSS animations/transitions, neutralizes sticky positions, shims IntersectionObserver, and tames requestAnimationFrame so the layout stays stable while capturing.

AI adapter (ai/ai_adapter.js): thin wrapper around Chrome built-in AI:

Summarizer for summaries

Rewriter to improve text

Writer for prompt-to-draft

LanguageModel (multimodal) to extract text from the last capture when no page text is available

PDF path: renders direct .pdf URLs with pdf.js and exports clean images.

Tech: Manifest V3, JavaScript/HTML/CSS, Canvas 2D, Chrome APIs (tabs, scripting, downloads).

Challenges we ran into

Seams & ghosting in stitched images → solved with consistent overlap cropping and a final canvas trim.

Dynamic pages (animations, sticky headers, lazy loading) → built a robust Freeze that pauses/neutralizes them.

Capture quotas / transient failures → added throttling, retries, and settle delays.

AI API warnings (outputLanguage) → language shims and removal of Proofreader to keep the UI warning-free.

Accomplishments that we’re proud of

Reliable, artifact-free full-page captures across tricky, real-world sites.

A tiny on-device AI panel that stays fast and private.

PDF Capture Mode that makes grabbing documents simple for reviewers.

Clean UX: progress messaging, preview link, one-click download, sensible fallbacks.

What we learned

Making stitching bulletproof is about small details: overlap size, timing, and last-slice clamping.

How to use Chrome’s built-in AI (Gemini Nano) effectively without any network dependency.

Robust content scripts matter: a good Freeze dramatically improves reliability.

What’s next for SnapSmart – Full-Page Capture + On-device AI

Smarter, per-site overlap heuristics and sticky-header detection.

Capture history, quick share/export presets.

More multimodal prompts (tables/figures extraction).

Optional text-only export (Markdown/PDF) alongside the stitched image.

Built With

  • ai
  • canvas-2d-chrome-apis:-tabs
  • chrome-extension-(manifest-v3)-javascript-(es)
  • downloads
  • html/css
  • languagemodel
  • multimodal)
  • pdf.js
  • pdfjs-dist)
  • rewriter
  • runtime
  • scripting
  • storage-chrome-built-in-ai:-summarizer
  • writer
Share this project:

Updates