Research map / Image generation / Fast diffusion sampling
MixGRPO: Unlocking Flow-based GRPO Efficiency with Mixed ODE-SDE
Junzhe Li, Yutao Cui, Tao Huang, Chuxuan Zeng, Weijie Kong, Yinping Ma, Chun Fan, Miles Yang, Zhao Zhong, Liefeng Bo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Global-SDE random-14-step optimization with mixed ODE-SDE sliding window (4 steps) | DanceGRPO on FLUX · HPDv2 (FLUX) | ImageReward: 1.436 -> 1.645 | Improved |
| replace Global-SDE with mixed ODE-SDE window | DanceGRPO on FLUX · HPDv2 (FLUX) | iteration time (s): 291.284 -> 149.326 | Improved |
| add DPM-Solver++ acceleration after the window (Flash*) | MixGRPO · HPDv2 (FLUX) | iteration time (s): 149.326 -> 83.278 | Improved |
| add DPM-Solver++ acceleration after the window (Flash*) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.645 -> 1.624 | No clear effect |
| use progressive window instead of random window (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.513 -> 1.629 | Improved |
| use progressive constant window instead of frozen window (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.580 -> 1.629 | Improved |
| set shift interval tau to 25 instead of 30 (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.589 -> 1.629 | Improved |
| set shift interval tau to 25 instead of 15 (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.509 -> 1.629 | Improved |
| increase window size from 1 to 4 (ablation) | MixGRPO · HPDv2 (FLUX) | Unified Reward: 3.235 -> 3.418 | Improved |
| increase window size from 4 to 6 (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.629 -> 1.547 | Hurt |
| use stride 1 instead of 2 (ablation) | MixGRPO · HPDv2 (FLUX) | ImageReward: 1.575 -> 1.629 | Improved |
| use second-order Midpoint solver instead of Heun (ablation) | MixGRPO-Flash · HPDv2 (FLUX) | ImageReward: 1.488 -> 1.578 | Improved |
Later papers that compared with this method
3 later accepted papers report MixGRPO as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- LeapAlign: Post-training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories (CVPR 2026 · better on 3 of 4 shared results)
- BranchGRPO: Stable and Efficient GRPO with Structured Branching in Diffusion Models (ICLR 2026 · better on 3 of 3 shared results)
- NoiseTilt: Noise-Tilted Reverse Kernels for Diffusion Reward Alignment (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TEMPFLOW-GRPO: WHEN TIMING MATTERS FOR GRPO IN FLOW MODELS (ICLR 2026 · similarity 0.91)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.90)
- DrPO: Drifting Preference Optimization for One-Step Generative Models (NeurIPS 2026 · similarity 0.87)
- TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment (NeurIPS 2026 · similarity 0.87)
- Value Gradient Guidance for Flow Matching Alignment (NeurIPS 2025 · similarity 0.87)
- Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
