✦ HAND MAGIC ✦

Made by Omar Musayev
STAR ON GITHUB
STYLE LAB

No styles yet.
Draw & save one!

DRAW YOUR STYLE

Write a few words — the model learns your stroke style.

GENERATE
STYLE
5.0

ABOUT HAND MAGIC — AI HANDWRITING GENERATOR

Hand Magic is a free, open-source AI handwriting generator that turns any text into realistic, human-like handwriting directly in your browser. Powered by two deep-learning architectures — a classic LSTM with Gaussian Mixture output (Graves 2013) and a modern cross-attention Transformer trained on the IAM On-Line Handwriting Database — Hand Magic produces handwritten text that varies stroke by stroke, just like a real person writing.

Use the LSTM mode to perform style transfer from your own handwriting: draw a sample on the canvas and the model learns your unique stroke style. Use the Transformer mode to generate multiple unique handwriting samples from a single prompt via top-k sampling. Everything runs server-side with no signup, no watermarks, no paywall.

The full source code, training pipeline, and pretrained model weights are open-source on GitHub under the MIT license.

FEATURES

  • Text-to-handwriting synthesis using LSTM and Transformer neural networks
  • Style transfer from user-drawn handwriting (LSTM mode)
  • Multi-sample generation with top-k sampling (Transformer mode)
  • Save, rename, and manage up to 10 handwriting styles per session
  • Asynchronous generation — samples stream as they finish
  • Two visual themes: minimal white and synth cyberpunk
  • No login, no API key, no signup required
  • Open source under the MIT license — fork, extend, self-host

FREQUENTLY ASKED QUESTIONS

Is Hand Magic really free?

Yes — completely free, no signup, no API key, no watermark. The code is MIT-licensed on GitHub.

Can the AI learn my own handwriting?

Yes. In LSTM mode, draw a few words on the canvas; the model uses your strokes to prime generation in your style.

Which AI models power Hand Magic?

A 3-layer LSTM (hidden size 400, 20-component Gaussian Mixture output) following Alex Graves (2013), and a 6-layer cross-attention Transformer (d_model 384, 6 heads) trained on IAM On-Line Handwriting DB.

Can I run Hand Magic on my own machine?

Yes — the entire pipeline (training, inference, FastAPI server, weights) is on GitHub. Clone the repo, install requirements.txt, and run uvicorn main:app.

What is the source code license?

MIT — you are free to use, modify, and redistribute the code with attribution.