✍️ Scripta

AI-Powered Handwriting Generation System

Transform plain text, PDFs, and DOCX files into realistic handwritten pages with modular rendering pipelines, writer-style simulation, and optional neural handwriting synthesis.

Generate believable handwritten pages using:

• Writer-style variation engine (WriterState)
• Baseline drift & spacing variation
• Ink simulation
• Natural imperfections
• Realistic paper rendering
• PNG & PDF export

Scripta supports:

• Plain text
• PDF documents
• DOCX files

Perfect for:

• assignments
• notes
• essays
• study material
• journals
• handwritten mockups

Choose from multiple writer personalities:

Experimental VATr++ integration enables:

• Neural handwriting synthesis
• Writer-conditioned generation
• Human-like stroke realism
• IAM handwriting dataset support
• Learned handwriting behavior

Input Parsing
      ↓
WriterState Variation Engine
      ↓
Glyph / Neural Rendering
      ↓
Page Composition
      ↓
Artifact Simulation
      ↓
PNG / PDF Export

scripta/
├── input_handler.py
├── glyph_store.py
├── variation_engine.py
├── renderer.py
├── page_compositor.py
├── neural_renderer.py
├── neural_page_compositor.py
└── artifact_sim.py

git clone <YOUR_REPO_URL>
cd Scripta

python3 -m venv .venv
source .venv/bin/activate

pip install --upgrade pip
pip install -r requirements.txt

mkdir -p fonts

curl -L https://raw.githubusercontent.com/googlefonts/caveat/main/fonts/ttf/Caveat-Regular.ttf \
-o fonts/Caveat-Regular.ttf

curl -L https://raw.githubusercontent.com/googlefonts/caveat/main/fonts/ttf/Caveat-Bold.ttf \
-o fonts/Caveat-Bold.ttf

python app.py

Open:

http://127.0.0.1:7860

python main.py \
  --input "Hello from Scripta" \
  --output output/hello.png

Optional configuration:

export GRADIO_SERVER_NAME=127.0.0.1
export GRADIO_SERVER_PORT=7860

Optional advanced setup for VATr++ neural handwriting synthesis.

git clone https://github.com/EDM-Research/VATr-pp.git VATr-pp

cd VATr-pp

pip install -r requirements.txt
pip install msgpack wandb

VATr-pp/files/vatrpp.pth
VATr-pp/files/resnet_18_pretrained.pth

data/iam/data_subset/data_subset/*.png

python scripts/prep_style_samples.py

python main.py \
  --backend neural \
  --input "Hello world" \
  --output output/neural.png

Automatically created when needed:

output/
fonts/
pages/
data/iam/

• Font backend works independently.
• Neural backend is optional.
• Missing assets produce friendly setup errors.
• Apple Silicon supported through PyTorch MPS.
• Designed for modular experimentation and extensibility.

• Font rendering pipeline
• UI rendering flow
• WriterState engine
• Paper + ink simulation
• PNG/PDF export

• VATr++ neural synthesis
• Neural writer cloning
• IAM style conditioning

Scripta aims to become a modular AI handwriting synthesis platform for:

• digital note generation
• educational tooling
• document realism pipelines
• synthetic handwriting research
• AI-assisted writing workflows