[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-grace-latent-compression-video-diffusion":3,"topics-all":38,"news-related-24d56562-4886-4169-8aa9-773f5fa34e2f":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"24d56562-4886-4169-8aa9-773f5fa34e2f","GRACE 把视频潜空间压 8 倍:提速 11 倍,跑分反超","KAIST AI 与 Kakao 提出 GRACE:冻结原编码器输出、另学残差潜变量并在 DiT 特征空间对齐,把 Wan2.1-14B 的潜 token 从 32.8k 压到 4.3k,延迟降 11.1 倍。VBench 文生视频 85.81 反超原版 83.93,图生视频持平,代码已开源。","480×832、81 帧的视频,让 Wan2.1-I2V-14B 在单张 A100 上端到端生成一遍,要 851.5 秒——超过 14 分钟。扩散模型的 DiT 要在 3.28 万个潜 token 上反复去噪 50 步,算力账就是这么烧起来的。KAIST AI 与 Kakao 的联合团队给出了一个新思路:不动扩散大模型,把它的\"燃料\"——潜空间——重新压缩一遍。方法叫 GRACE,论文已挂 arXiv(2610.10524),代码以 MIT 协议开源,权重放在 Hugging Face。\n\n## 压缩潜空间的老大难\n\n视频扩散模型的速度瓶颈在 token 数:自编码器压缩率越高,DiT 要处理的 token 越少,生成越快。但压缩率一旦拉高,重建质量会先崩——想靠加通道数找回细节,更多通道又会拖慢 DiT 收敛;更麻烦的是,压缩后的潜分布偏离了 DiT 预训练时学到的分布,要么从头重训 DiT,要么付出高昂的适配成本。只对着重建指标优化编码器,表面看兼容,实际上潜空间已经被悄悄搬离了 DiT 认识的世界。\n\n## GRACE 的三件套\n\nGRACE 分两阶段绕开这些坑。第一阶段训自编码器:冻结原版 Wan2.1 编码器输出的基座潜变量,另外学一个残差潜变量,装下强压缩丢掉的细节和运动信息;损失函数里,压缩潜变量会被拉到冻结 DiT 的特征空间里与原版潜变量对齐,让编码器\"为生成而优化\"。第二阶段适配 DiT:输入输出投影扩展到残差通道,transformer 块上挂 LoRA,再加上非对称去噪——基座潜变量永远比残差领先一步去噪(默认偏移 δ=0.15),残差始终建立在更\"安定\"的底子上,且不增加采样成本。最终潜配置从 f8t4p2 压到 f16t8p2,480×832 下的 token 从 32.8k 降到 4.3k。\n\n## 快了 11 倍,文生视频跑分反超\n\n结果表最有说服力:同样 480×832×81,GRACE 端到端 75.8 秒,原版 Wan2.1-14B 要 851.5 秒,提速 11.1 倍;分辨率拉到 736×1280×81,提速放大到 15.5 倍——分辨率越高省得越多。质量侧,VBench 文生视频总分 GRACE 85.81,反超未压缩原版的 83.93,其中语义分 84.98 对 78.70 拉开最大差距;图生视频总分 87.90,与原版 87.92 基本持平。横向比,同 token 量级的 Open-Sora 2.0、DC-Gen、LTX-Video 在文生视频总分上都不及 GRACE,图生视频总分 LTX-Video 以 87.97 与 GRACE 处在同一档。\n\n## 消融:预训练一次都别浪费\n\n论文的消融实验把\"为什么不能蛮干\"讲得很清楚:同样压缩、把 DiT 从头重训同样步数,VBench 文生视频总分只剩 71.04,比\"只压缩不对齐\"的 baseline 82.12 还低 11 分;依次加上双潜变量设计(83.55)、DiT 特征空间对齐(84.68)、非对称去噪(85.81),分数一级级爬回并反超原版。对齐目标也有讲究:换成 V-JEPA 2.1 的特征只有 83.99,不如冻结 DiT 自己的特征——生成任务里,DiT 自己的\"世界观\"更值钱。去噪顺序同样有方向性:基座先行 87.86,残差先行掉到 87.17,低于同步去噪的 87.51。\n\n这套方法证明:预训练视频扩散模型的算力账,可以从潜空间这一层重新算——不重训大模型,token 少一个数量级,质量守得住。视频生成成本高企的当下,\"生成感知\"的压缩很可能变成标配工序。代码和权重都已放出,复现门槛不高,值得动手一试。\n\n参考:arXiv:2610.10524 \u002F cvlab-kaist.github.io\u002FGRACE","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.10524","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"7b67033c-19e6-4052-a626-e681bba64c7a","diffusion",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",{"id":22,"name":23,"slug":23,"description":14,"color":14},"ebe5dcd1-46b1-4298-b8c2-8e0e2f456e56","video-generation",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"380b09db-7d54-40d9-845f-d136f533646c","en","GRACE: 8x Fewer Latent Tokens, 11x Faster Video Diffusion","GRACE compresses Wan2.1-I2V-14B latents 8x: 11.1x faster at 480x832, 15.5x at 736x1280, VBench T2V total 85.81 beats the uncompressed 83.93.","Generating a single 480x832, 81-frame video with Wan2.1-I2V-14B takes 851.5 seconds end to end on one A100 — more than 14 minutes. The Diffusion Transformer denoises roughly 32.8k latent tokens over 50 steps, and that is where the compute bill comes from. A joint team from KAIST AI and Kakao proposes a different way to save: leave the pretrained diffusion model alone, and recompress its fuel — the latent space. The method, GRACE (Generation-Aware Latent Compression for Efficient Video Generation), is on arXiv (2610.10524), with code released under MIT and weights on Hugging Face.\n\n## Why compressing latents is hard\n\nThe speed bottleneck of video diffusion is token count: the higher the autoencoder's compression ratio, the fewer tokens the DiT processes, and the faster generation gets. But push compression too far and reconstruction quality collapses first — recovering detail requires more channels, which slows DiT convergence. Worse, the compressed latent distribution drifts away from what the DiT learned during pretraining, so the DiT must be retrained from scratch or adapted at considerable cost. Optimizing the encoder for reconstruction alone looks compatible on paper, but silently moves the latent space out of the world the DiT knows.\n\n## The three pieces of GRACE\n\nGRACE works in two stages. Stage 1 trains the autoencoder: a frozen base latent from the original Wan2.1 encoder anchors the representation, while a learned residual latent carries the detail and motion that stronger compression throws away. Beyond reconstruction losses, the compressed latent is aligned with the pretrained one inside the frozen DiT's feature space, so the encoder is optimized for generation rather than reconstruction alone. Stage 2 adapts the DiT: input and output projections are extended to the residual channels, LoRA is attached to the transformer blocks, and asymmetric denoising keeps the base latent denoised slightly ahead of the residual (default offset delta = 0.15), so the residual always builds on a more settled base, at no extra sampling cost. The latent configuration goes from f8t4p2 to f16t8p2, and tokens at 480x832 drop from 32.8k to 4.3k.\n\n## 11x faster, and T2V scores go up\n\nThe results table is the persuasive part: at 480x832x81, GRACE takes 75.8 seconds end to end versus 851.5 seconds for the original Wan2.1-14B — an 11.1x speedup — and at 736x1280x81 the speedup grows to 15.5x. On quality, GRACE scores 85.81 on VBench text-to-video total, beating the uncompressed original's 83.93, with the semantic score gap the widest (84.98 vs 78.70); image-to-video total is 87.90 versus 87.92 for the original, essentially flat. Among methods at the same token budget — Open-Sora 2.0, DC-Gen, LTX-Video — none matches GRACE's T2V total, while LTX-Video's I2V total (87.97) sits in the same tier.\n\n## Ablations: never waste the pretraining\n\nThe ablation study makes clear why brute force fails: with the same compression, training the DiT from scratch for the same number of steps yields a VBench T2V total of just 71.04 — 11 points below the 82.12 of the compress-only baseline. Adding the dual-latent design (83.55), the DiT-feature-space alignment (84.68), and asymmetric denoising (85.81) climbs the score back above the original. The alignment target matters too: video foundation model V-JEPA 2.1 features reach only 83.99, below the frozen DiT's own features at 84.68 — in generation tasks, the DiT's own worldview is worth more. Denoising order is directional as well: base-first scores 87.86 on I2V total, while residual-first drops to 87.17, below even the synchronous schedule's 87.51.\n\nThe significance goes beyond a single number: the compute bill of a pretrained video diffusion model can be renegotiated at the latent layer — no DiT retraining, an order of magnitude fewer tokens, and quality holds. As video generation costs stay high, generation-aware compression may well become a standard step in the pipeline. Code and weights are out, reproduction cost is low, and it is worth trying hands-on.\n\nReference: arXiv:2610.10524 \u002F cvlab-kaist.github.io\u002FGRACE","grace-latent-compression-video-diffusion","2026-10-08T15:10:59Z","2026-10-08T15:11:18.024629Z","2026-10-08T15:11:18.024644Z",true,"agent",199,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":35,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"caed836e-2168-418f-b5c1-bde3ce962e66","Mask Forcing 往蒸馏 rollout 里掺干净 token:修视频生成的模式坍缩,指令遵循最高涨 6.5 分","mask-forcing-video-diffusion-distillation","2026-09-09T23:08:37+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"02c8b500-ec11-44a6-8c58-6e880563dad8","FastH3 开源:4 步蒸馏版 MiniMax H3,B200 单卡最高提速 14 倍","fasth3-4-step-distilled-minimax-h3","2026-08-30T21:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"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":75,"title":76,"news_slug":77,"published_at":78},"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","tencent-hunyuan-meanflownft","2026-07-16T12:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"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":85,"title":86,"news_slug":87,"published_at":88},"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"]