Rectified Flow matching, optimal transport vector fields, and Patchified Diffusion Transformers (Sora/Flux).
For a small example, a noisy scalar should travel toward a target value. Define an interpolation path, compute its target velocity, and integrate a learned approximation over several time steps.
A diffusion transformer learns latent image trajectories. Patch latent inputs, condition on time and text, train velocity prediction, and compare sampler steps, guidance, and fidelity.
Initializing Interactive Playground...
Predict how one control changes the Flow Matching & DiT Video result.
The starter deliberately follows a tempting but flawed assumption: Confusing the training path with the sampled trajectory
Check: Your saved hypothesis predicts a direction, the experiment tests it, and the result changes or supports your explanation.
Explanations after each answer · 80% to pass · Flow Matching & DiT Video
Chapter 1 of 6 · Core
Classical generative diffusion models (DDPM/DDIM) construct probability paths by gradually adding Gaussian noise to clean data via a forward Stochastic Differential Equation (SDE):
pushing a simple base noise distribution at to the complex data distribution at .
Key equations
Two free references selected for Flow Matching & DiT Video.
Primary sources and free technical reading for Flow Matching & DiT Video.