MoDDiT: Simple Token Routing in Diffusion Transformers
We propose MoDDiT, a simple architecture modification that learns to route tokens through both MLP and self-attention operations.
ProofPaper ↗
Key points
- Real-time video generation, interactive world models, and on-device image synthesis are almost a reality.
- Diffusion Transformers (DiT) achieve excellent sample quality, shifting the bottleneck to the throughput of the models themselves.
- Using MoDDiT to train image generation models from scratch significantly reduces computational costs of training and inference, processing as few as 12.5% of the tokens.
- We also use MoDDiT to train video generation models, where allocation correlates with motion, and to distill a dense bidirectional video model into a sparse autoregressive student.
Sources (1)
- [1]MoDDiT: Simple Token Routing in Diffusion TransformersarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:05 PM
We propose MoDDiT, a simple architecture modification that learns to route tokens through both MLP and self-attention operations.
Real-time video generation, interactive world models, and on-device image synthesis are almost a reality.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Do Diffusion LLMs Scale? Testing Mercury 3 + Decide
- Oct 20, 2026Learning the 4D World through Point Trajectory Forecasting
- Oct 20, 2026OverSeer: Multi-Agent World Models Beyond What Agents See
- Oct 19, 2026SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation
- Oct 19, 2026SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration
- Oct 19, 2026ActionRoPE: action as coordinates for spatially consistent control in planar-camera game world models