Project library
Synthetic Digit Generator
Generate new handwritten-style digits and explore how a model learns their shapes.
STAR Summary
- Situation
- I wanted to generate new handwritten digits, not just recognize them.
- Task
- Train a generative model that can produce new MNIST-style digits from a learned latent space.
- Action
-
- Trained a Variational Autoencoder (VAE) on MNIST (60,000 training digits) with a 20-dim latent space for up to 100 epochs.
- Result
-
- Generated new digits by sampling the learned latent space.
- Saved the trained model so generation is a quick inference step.
Digit Generator
Choose a digit and generate a collection of handwritten variations.
Open full demo
Generate a grid of synthetic digits and explore how the output morphs.
- Choose Auto or a digit from 0–9 with the Digit control.
- Click “Generate” to refresh the grid.
- Open “Advanced settings” below the grid to adjust grid size, seed, latent dimension, and distortion.
- Generate again to compare how settings change the outputs.
