MNIST-Style Digit Generator

Generate handwritten digits using Latent Flow Matching!

Background: Using a standard VAE and a MLP for the velocity predictions, we can create a rudimentary digit generator that produces new images around the original MNIST image distrubution. This servers as an accessible interface to run the model without setting up and strictly running the original source files while also keeping low compute (instead of fully running an overkill MMDiT).

Illustration of the MNIST digit generation model

Here is 4x4 example of generated numbers after 100 timesteps.

Choose a digit: 5 Batch size: 4 Total Steps: 100

When send some conditioning, we may also send a "null" input that produces random images.
Preferably, send a batch size that is a power of 2 for efficient kernal calculations

Generation Process

Watch the digit emerge from random noise.

Animated generation of a handwritten digit

Your animation will appear here.

Generation progress

Step 0 / 100

Final Result

Your final digit will appear here.