moebiusT7/book-ocr-studio
0
1<!doctype html>2<html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">3<title>Book OCR Studio — Local-first</title>4<style>5:root{color-scheme:light}*{box-sizing:border-box}body{margin:0;background:#f5f4ef;color:#18342f;font:17px/1.65 system-ui,sans-serif}main{max-width:1000px;margin:auto;padding:70px 26px}small{letter-spacing:.12em;text-transform:uppercase}h1{font-size:clamp(38px,7vw,70px);line-height:1.08;letter-spacing:-.05em;margin:20px 0}h2{font-size:24px}p{max-width:720px}.lead{font-size:23px;color:#426059}.badge{display:inline-block;background:#e2eadf;border-radius:24px;padding:6px 14px;font-size:14px}.button{display:inline-block;background:#215947;color:white;text-decoration:none;border-radius:8px;padding:13px 22px;margin:12px 8px 12px 0}.secondary{background:white;color:#215947;border:1px solid #baccc1}.grid{display:grid;grid-template-columns:repeat(3,1fr);gap:18px;margin:40px 0}.card{background:white;padding:24px;border-radius:14px;border:1px solid #e0e5df}.card h2{margin-top:0}a{color:#215947}code{background:#e8ebe3;padding:3px 7px;border-radius:4px}footer{margin-top:50px;border-top:1px solid #ccd6cd;padding-top:22px;font-size:14px;color:#53645c}.shots{display:grid;grid-template-columns:1fr 1fr;gap:22px;margin:34px 0 10px}.shots figure{margin:0}.shots .wide{grid-column:1/-1}.shots img{width:100%;height:auto;border-radius:10px;border:1px solid #d9dfd8;background:white}.shots figcaption{font-size:15px;color:#53645c;margin-top:8px;max-width:none}.shots figcaption small{display:block;font-size:13px;color:#6b7a73;margin-top:4px}.hint{font-size:14px;color:#6b7a73;margin:-6px 0 0}.prov{font-size:13px;color:#6b7a73;max-width:none}@media(max-width:650px){main{padding-top:36px}.grid,.shots{grid-template-columns:1fr}}6</style></head><body><main>7<small>Linux source beta · Runs on your computer</small>8<h1>Your books.<br>Your local workflow.</h1>9<p class="lead">Your Kindle books, PDFs and page images → local OCR and Gemma review → Markdown, HTML, PDF and EPUB.</p>10<span class="badge">Local-first · Source language preserved</span>11<p><a class="button" href="book-ocr-studio-source.zip" download>Download source ZIP</a><a class="button secondary" href="install.html">Installation guide</a></p>12<p>This is a distribution page, not an online OCR service. Nothing uploaded here is processed. Install the application locally with an NVIDIA GPU and Ollama; the standard setup downloads Gemma 12B locally. C1 OCR review is included; 26B is optional.</p>13<h2>What it looks like</h2>14<p class="hint">Click any image to enlarge it.</p>15<div class="shots">16<figure><a href="docs/images/01-start.png"><img src="docs/images/01-start.png" alt="Start screen with PDF / Images and Kindle capture tabs, page range, GPU mode and the Start conversion button" loading="lazy"></a><figcaption><b>1 · Start.</b> Upload a PDF or page images, or switch to the Kindle capture tab.</figcaption></figure>17<figure class="wide"><a href="docs/images/02-review.jpg"><img src="docs/images/02-review.jpg" alt="Review screen: source page image beside Original OCR, Gemma suggestions and Changes tabs, with Approve and Restore buttons" loading="lazy"></a><figcaption><b>2 · Review against the page image.</b> Compare the page image with the suggested text; the original OCR is kept until you approve a page.<small>On this page Gemma proposed "cannot" for "can neither". The image check did not support it, so the suggestion was deferred and the original reading kept — see the note above the Approve / Restore buttons.</small></figcaption></figure>18<figure class="wide"><a href="docs/images/03-markdown.png"><img src="docs/images/03-markdown.png" alt="Exported Markdown with source-page markers, and a question answered from it by a local model" loading="lazy"></a><figcaption><b>3 · Use the Markdown.</b> The export keeps a source-page marker per screen, so an LLM's answer can be checked against the page.</figcaption></figure>19</div>20<p class="prov">Image 2 shows a public-domain scan (Ruskin, <i>The Stones of Venice</i>, 1890 edition, page 50, via Wikimedia Commons); images 1 and 3 use a short synthetic text written for this documentation. The answer in image 3 came from the local gemma4:12b through Ollama. The images illustrate the interface, not an accuracy level.</p>21<div class="grid">22<section class="card"><h2>Import</h2><p>Use PDFs or page images, or connect a supported Kindle reader in Chrome. Source material stays local in the default Gemma workflow.</p></section>23<section class="card"><h2>Review</h2><p>Gemma 12B is the default; 26B is optional. Möbius Custom C1 for OCR is an application review profile, not new model weights.</p></section>24<section class="card"><h2>Export</h2><p>Download Markdown for model context, or create HTML, PDF and EPUB separately. Original OCR and unapproved suggestions remain distinct.</p></section>25</div>26<h2>Start locally</h2><p>Extract the ZIP, read the installation guide, then run <code>bash install.sh</code> and <code>bash run.sh</code>. Install Ollama first. Python 3.10/3.11 and compatible NVIDIA drivers are required. The installer prepares 12B; see the guide for Chrome and optional 26B setup.</p>27<p><a href="benchmarks.html">12B / 26B comparison and limitations</a></p>28<h2>Optional vision models</h2><p>Gemma 4 remains recommended. An opt-in OpenAI-compatible connector lets you choose another vision-capable model. Enabling it sends images and OCR text to your chosen endpoint. API compatibility does not guarantee completion or accuracy. Read the <a href="connectors.html">connector setup and privacy notes</a>.</p>29<h2>Know the limits</h2><p>OCR and model suggestions can be wrong. Check important names, numbers and quotations against source images. GPU compatibility, languages and browser layouts are not universally validated. Read <a href="release-notes.html">the validation scope</a> before relying on this beta.</p>30<footer><a href="license.html">AGPL-3.0-only application source</a> · <a href="licenses.html">Third-party and model terms</a><p>No model weights, books or credentials are included. Process only content you are entitled to use. Not affiliated with Amazon.</p></footer>31</main></body></html>32