meanflow

rCM & Causal-rCM: Leading and Unified Algorithms/Infrastructures for Bidirectional/Autoregressive Video Diffusion Distillation at Scale

[arXiv 2026] This is the official PyTorch implementation of "MeanFlowNFT: Bringing Forward-Process RL to Average-Velocity Generators".

[RSS 2026] LPS: Latent Policy Steering through One-Step Flow Policies.

MLX package for the dots TTS model, with fused attention and meanflow

Unofficial PyTorch implementation of "Mean Flows for One-step Generative Modeling" (2025). Extends the framework to Edge-to-Image generation with Multi-scale Edge Encoder and 1-NFE sampling.

Interactive dependency-free lab on flow matching, rectified flow & MeanFlow one-step generation

DDPM, DDIM, GAN and MeanFlow implemented from scratch and compared under optimal-transport metrics — DDIM at 100 steps beats DDPM at 1000, and MeanFlow reaches competitive quality in a single network evaluation.

An implementation of Flow Matching based models