Diffusion Explorer is an interactive tool (check out a beta version here) for communicating the geometric intuitions behind diffusion and flow based generative models. This project is currently a work in progress.

Diffusion Explorer is mainly an educational tool with the following key functionality:

1. Implements various training objectives like Flow Matching and Denoising Score Matching
2. Shows the dynamics of generated samples over time for pretrained models
3. Allows a user to observe how generated samples change through training
4. Enables training on custom hand drawn distributions

An interactive blog post explaining Rectified Flow, a technique for straightening the trajectories learned by flow matching models. The explainer features animated visualizations that demonstrate:

• How flow matching learns curved trajectories between distributions
• Why curved paths are problematic for few-step sampling
• How rectified flow iteratively straightens trajectories
• The connection between straight paths and optimal transport

Live demo: alechelbling.com/blog/rectified-flow

cd diffusion-explorer/apps/rectified-flow-explainer
npm install
npm run dev

Then open your browser to the URL shown in the terminal (typically http://localhost:5173).

To retrain the flow matching and rectified flow models:

npm run train:flow-matching    # Train base flow matching model
npm run train:rectified-flow   # Train rectified flow model (3 steps)
npm run cache-samples          # Regenerate cached trajectories

You can try out Diffusion Explorer locally by running the project.

First, clone the project:

git clone https://github.com/helblazer811/Diffusion-Explorer

Then change the directory and install dependencies:

cd diffusion-explorer
npm install

Now run the local server:

npm run dev

and then access it in your browser at the specified port.

You can also see some other interesting (non-interactive) visualizations in /other-visualizations.