← All threads

Inference-Time Scaling for Diffusion

Investigates methods for scaling inference compute in diffusion and flow-matching generative models to improve generation quality.

5 papers

Where this stands

The written synthesis of this thread is for subscribers. Subscribe.

How this thread developed

  1. Oct 2026 · Shanghai Jiao Tong University, Alibaba Group

    RL-trained cache scheduler adapts video diffusion to user-specified speedups

    Introduces RL-optimized cache scheduling for video diffusion to achieve user-specified speedups, focusing on terminal error.

  2. 2 further papers

    Oct 2026 · Warsaw University of Technology, IDEAS Research Institute

    Recursive latent reasoner boosts diffusion models for visual puzzles

    Adds a small recursive reasoner that refines latent states within each denoising step, scaling inference-time compute for diffusion models.

  3. Oct 2026 · University of Geneva, Meta FAIR

    Velocity scaling improves flow models by correcting sampling time lag

    Introduces velocity scaling correction that improves flow model sample quality by addressing sampling time lag.

3 of 5 papers shown