Text to Handwriting AI

Turn typed text into realistic handwritten pages, or build a complete handwritten assignment, with a filled-in front page and AI-drafted answers.

Maintained by CodewithWitcher. Contributions are welcome. See CONTRIBUTING.md.

Use it

Your data stays in your browser in every option. The online version calls AI providers straight from your browser. Claude and DeepSeek allow that, but some providers block it (CORS). If a provider fails online, use the one-command option: it runs a small local relay that works with all of them.

Two modes

Use the switch at the top of the page.

Classic

The original quick converter. Type or paste text, choose a font, ink and paper, then generate images or a PDF.

Advanced (AI)

A guided, four-step flow:

1. Template & handwriting. Pick an assessment template (Q&A, short answers, detailed answers, essay, lab report, case study, or your own). Choose a font or upload your own .ttf, then set ink colours for answers and headings, heading style, ruled lines, margin, and natural variation (small per-word tilt so it doesn't look like a font). A live preview updates as you change things.

2. Front page (optional). Upload your university or college front page (image or PDF). Auto-detect uses OCR to find blanks such as "Name:", "Roll No.:" and "Subject:", then writes your details in your handwriting. Smart detect (AI) lets the AI decide which blank is which. Drag any value to fine-tune it, and save the layout to reuse next time.

3. Questions & answers. Choose one of three sources:

   • Upload the question paper (PDF). The AI lists every question with its marks. Scanned PDFs are read with OCR, or directly by Claude.
   • Paste the questions.
   • Write or paste your own content. No AI is needed. Use simple formatting for headings and lists.

   Then generate the answers one by one with AI. Answer length follows the marks and the template. You can edit, regenerate or reorder anything. No API key? Click Copy prompt, paste it into ChatGPT, Claude.ai or DeepSeek, paste the reply back, and click Import answers.

4. Preview & download. Text flows across A4 pages: words are never cut, and headings are never left alone at the bottom of a page. Edit the final text if you like, then download one PDF (front page first) or the individual page images.

My Profile stores your name, roll number, course, subject and other details in this browser. To fill it from a file, click Download profile template, fill in the JSON in any text editor, then Import profile. AI Settings supports Claude, DeepSeek, OpenAI, Google Gemini, OpenRouter, Groq, Ollama (local, no key) and any OpenAI-compatible endpoint.

Run locally

One command, no clone. Use whichever you have:

pipx run text-to-handwriting-ai         # Python (pipx)
uvx text-to-handwriting-ai              # Python (uv)
npx text-to-handwriting-ai              # Node.js

Each one opens http://127.0.0.1:8080 in your browser. Options: --port 9000, --no-open, --host 0.0.0.0 (to reach it from other devices on your network).

To keep it installed instead: pip install text-to-handwriting-ai, then run text-to-handwriting-ai.

From a clone (no dependencies needed to run):

npm start              # Node 18+
python3 server.py      # or Python 3.8+, same behaviour

Both serve the app and relay AI requests to the provider you choose, so browsers don't block them (CORS). If the port is busy, the next free one is used.

macOS: port 5000 is taken by AirPlay Receiver, which is why the default is 8080.

Privacy

• Everything you enter (profile, templates, fonts, front pages, drafts, API keys) is stored only in this browser (localStorage and IndexedDB).
• API keys are sent only to the provider you select. When the local server relays a request, it doesn't store or log keys or content.
• Only academic context (course, subject, semester) is included in AI prompts. Your name and ID numbers are never sent.
• My Profile → Delete all saved data wipes everything. Data exports never include API keys.

Content formatting

Own content and the final text step use this formatting:

Project structure

index.html              Page shell: Classic markup + Advanced markup
bin/cli.mjs             `npx text-to-handwriting-ai` launcher (Node, no deps)
text_to_handwriting_ai/ Same server + AI relay in Python (stdlib only); the pip package
server.py               Runs the Python server from a clone
pyproject.toml          pip package definition (bundles the site files at build time)
js/app.mjs              Classic mode (original behaviour)
js/mode-switch.mjs      Classic / Advanced switch (lazy-loads Advanced)
js/config.mjs           Repository / author links
js/advanced/
  main.mjs              Advanced mode controller (tabs, steps)
  llm.mjs               AI providers (Claude SDK + OpenAI-compatible)
  questions.mjs         Question extraction (AI and rule-based)
  templates.mjs         Templates, prompts, document composition
  renderer.mjs          Pagination + handwriting page rendering
  frontpage.mjs         Front page detection / editor
  pdf-input.mjs         PDF text extraction + OCR
  profile.mjs, store.mjs, output.mjs, ui.mjs, loaders.mjs
css/advanced.css        Advanced mode styles

Libraries

• html2canvas: renders pages to images
• jsPDF: builds the PDF
• pdf.js: reads uploaded PDFs
• Tesseract.js: OCR for front pages and scanned papers
• Anthropic TypeScript SDK: Claude integration
• Cypress: end-to-end tests (npm test, with the server running)

Contributing

Bug reports, front-page layouts that don't detect well, new templates and fixes are all welcome. Read CONTRIBUTING.md to get set up, and follow our Code of Conduct. To report a security problem, see SECURITY.md.

Responsible use

Use this for notes, practice, accessibility (for example, if handwriting is hard for you) and mock-ups. Check your institution's rules before submitting AI-assisted or generated work.

License

MIT. See LICENSE. Based on the MIT-licensed text-to-handwriting project.