[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-tencent-hunyuan-meanflownft":3,"news-related-0599b775-ac17-49d2-aebd-a16f531c7168":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"0599b775-ac17-49d2-aebd-a16f531c7168","腾讯混元 MeanFlowNFT：把 RL 接进「平均速度生成器」，Wan 2.1 4 步反超 50 步 LongCat-Video RL","短视频模型的开销正在被「少步生成」路线重塑。DiffusionNFT 用前向过程的负似然估计做 RL 后训练,避开了反向轨迹和似然估计的高昂成本,但它原本针对的是「瞬时速度」的扩散模型;而 MeanFlow 这种「平均速度」生成器走的是另一条路——靠预测一段时间区间上的平均速度,把采样步数压到 4 步甚至 1 步。两套公式不同,RL 优化目标一直接不上。腾讯混元 7 月 16 日公开的 MeanFlowNFT(arXiv 2607.15273)把这个缺口补上了。团队用 MeanFlow identity 在「平均速度」和「瞬时速度」之间建了一座桥:训练时构造一个 induced 瞬时速度预测器,把 DiffusionNFT 的目标函数套上去;采样阶段仍然只跑平均速度,所以原本的少步生成优势一个字符不掉。论文还证明 MeanFlowNFT 继承了 DiffusionNFT 的策略单调改进保证。数字最有说服力:Wan 2.1 视频模型上,4 步 MeanFlowNFT 把 VBench 推到 84.33,直接压过 50 步的 LongCat-Video RL(82.57);SD3.5-M 图像生成上,8 个评估维度里 6 个超过此前所有 RL 调优的少步生成器。项目已开源(github.com\u002FHarahan\u002FMeanFlowNFT)。把 RL 后训练的成本曲线从「必须配合多步 ODE solver」改成「少步也能稳定对齐奖励」,对实时视频生成、交互式视频 Agent 这些场景是直接利好——这篇工作值得 Hunyuan 之外的从业者留意:它不是单一模型的升级,而是给整个少步生成路线补上 RL 这一课。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.15273","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":18,"name":19,"slug":19,"description":13,"color":13},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":21,"name":22,"slug":22,"description":13,"color":13},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":28},"fefd0b76-9220-4485-8aac-bdf25731faa2","en","MeanFlowNFT: RL in mean-flow generators, 4 steps beat 50","The cost of short-video models is being reshaped by the \"few-step generation\" route. DiffusionNFT uses the forward-process negative likelihood estimate for RL post-training, avoiding the high cost of reverse trajectories and likelihood estimates, but it was originally designed for \"instantaneous-velocity\" diffusion models; MeanFlow, the \"average-velocity\" generator, goes a different way — it predicts the average velocity over a time interval, compressing sampling steps to 4 or even 1. The two formulas differ, and the RL optimization target had no way to connect. Tencent Hunyuan's MeanFlowNFT (arXiv 2607.15273), released July 16, fills this gap. The team uses the MeanFlow identity to build a bridge between \"average velocity\" and \"instantaneous velocity\": in training, an induced instantaneous-velocity predictor is constructed and the DiffusionNFT objective is applied; at sampling, the model still runs only average-velocity, so the few-step generation advantage is preserved to the letter. The paper also proves MeanFlowNFT inherits DiffusionNFT's policy-monotonic-improvement guarantee. The numbers are most persuasive: on the Wan 2.1 video model, 4-step MeanFlowNFT pushes VBench to 84.33, directly beating 50-step LongCat-Video RL (82.57); on SD3.5-M image generation, it beats all previous RL-tuned few-step generators in 6 of 8 evaluation dimensions. The project is open-sourced (github.com\u002FHarahan\u002FMeanFlowNFT). Changing the RL post-training cost curve from \"must be paired with a multi-step ODE solver\" to \"few steps can stably align rewards\" is a direct boon to real-time video generation, interactive video Agents, and the like. This work deserves attention from practitioners beyond Hunyuan: it's not a single-model upgrade, but a course correction for the entire few-step generation route, completing the RL chapter.","tencent-hunyuan-meanflownft","2026-07-16T12:00:00Z","2026-07-19T00:12:29.890817Z","2026-08-19T02:08:40.142862Z",true,"agent",134,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"18d2aa73-7244-4b10-b611-46475e17327e","ForgeWM开源:一步去噪72FPS的可玩世界模型,8张卡复现全流程","forgewm-few-step-playable-world-model","2026-08-24T21:10:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"42cfc778-8f1b-4bf2-a0ae-4343a066f48d","RhymeFlow：清华提出异步去噪流调度，DiT视频生成训练免费加速1.53倍","rhymeflow-tsinghua-async-denoising-1-53x","2026-06-07T22:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"bf8755fc-cd4f-4bd9-9617-e70f56ddc4ac","LTX-2.3：开源视频生成正式进入 4K + 原生音频时代","ltx-2-3-lightricks-4k-native-audio","2026-06-02T01:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"2874a2e5-beae-4627-8f6f-a34cf2cc8d7a","一段随手拍视频直出4D人体:4DAnyone用RCP+TCR破解多视角一致性,代码权重全开源","4danyone-monocular-video-4d-human","2026-08-20T17:59:53+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"5612d186-46ee-4509-9a93-94045ba004ae","LTX-2.5 开放权重视频模型:4K 反而在 Fast 端点,EXR 色彩管线也焊进去了","ltx-2-5-open-weights-video","2026-08-18T15:20:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"6f9e9f94-9dcc-4c6c-b254-6c5d0fe8ed37","京东开源 JoyAI-Video-Edit:16B 多模态扩散 Transformer 把视频编辑推进「边播边改」实时流时代","jd-joyai-video-edit-realtime-diffusion","2026-08-10T00:00:00+00:00"]