MeanFlowNFT: Forward Process RL for Mean Velocity Generators
Tencent-Hunyuan · hf · 2026-07-17
To apply DiffusionNFT—an efficient reinforcement learning framework that doesn't require reverse trajectories—to MeanFlow generators, this paper proposes MeanFlowNFT.
- Bridging the Gap: DiffusionNFT optimizes instantaneous velocity, while MeanFlow relies on average velocity. This method leverages the MeanFlow identity to construct an induced instantaneous velocity predictor, making reward optimization well-defined in MeanFlow.
- Theoretical Guarantee: Preserves the strict policy improvement guarantee of DiffusionNFT without breaking MeanFlow's fast, few-step generation characteristics.
- Excellent Results: Consistently improves baselines on image and video generation. On Wan 2.1, the 4-step MeanFlowNFT's VBench score (84.33) even surpasses the 50-step LongCat-Video RL (82.57).
More from Multimodal
- Seedance 2.0 demo turns ketchup on spaghetti in Rome into an AI reaction meme — azed_ai · 2026-07-21
- A reusable “Lunar Eclipse Dreamscape” prompt comes with multiple example renders — LudovicCreator · 2026-07-21
- Midjourney 8.2 preview shows a double-exposure prompt with strong style control — michaelrabone · 2026-07-21
- Travel MCP Server adds flight, hotel, weather and budget tools for agents — modelcontextprotocol · 2026-07-21
- Douyin Video Analysis MCP turns share links into structured video summaries — modelcontextprotocol · 2026-07-21
- Synthesia launches Dubbing 2.0 with 130+ languages and lip-sync video translation — synthesiaIO · 2026-07-21