[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-kling-4-0-30s-4k-video":3,"topics-all":38,"news-related-75e95a3e-4574-4a61-8785-9e13b599aa02":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},"75e95a3e-4574-4a61-8785-9e13b599aa02","可灵 Kling 4.0:30 秒 4K 视频一次直出","快手可灵发布 Kling 4.0：单次最长生成 30 秒原生视频，支持 10 张关键帧控制叙事节点，15 项多模态参考（10 图+5 视频+7 主体），输出 4K 10-bit HDR 与 21:9 画幅，提示词上限 8000 tokens，续拍延展至 2 分钟。Flash 版已开放会员体验，完整版 10 月上线。","视频生成模型的「短视频天花板」被抬高了：单次生成时长从行业普遍的几秒量级，一步跨到 30 秒原生直出。快手可灵 9 月 28 日向黑金年卡会员开放 Kling 4.0 Flash 抢先体验，完整版定档 10 月正式上线——单次 30 秒，是这代升级里最硬的一个数字（[IT之家报道全文](https:\u002F\u002Fwww.ithome.com\u002F1\u002F008\u002F073.htm)）。\n\n## 30 秒直出，配一整套专业规格\n\n官方公告的数字值得逐条看：单次生成视频最长 30 秒，支持长镜头与连续叙事；输出规格覆盖 4K、1080p 的 10-bit HDR（官方标注「即将上线」）与 21:9 超宽画幅；提示词输入上限提升到 8000 tokens，模型还会对提示词做补充、扩写与润色。音频侧支持高品质双通道立体声，角色口型匹配精度进一步打磨，官方称之为更接近同期声的听感。\n\n对专业创作者来说，10-bit HDR 加 21:9 的意义不只是「更清楚」：官方口径是逆光不爆、夜景不糊、霓虹不晕，给后期调色留出更宽裕的空间——这是往影视工业交付标准靠拢的信号，而不只是消费级玩法的堆料。\n\n## 从「抽卡」到「导演」：可控性才是主战场\n\n如果说上一代视频模型卷的是画质，Kling 4.0 明显把重心压在了可控性上，官方列出的能力清单几乎是一份「导演控制台」：\n\n- **多关键帧控制**：最多 10 张关键帧输入，自由设定画面节点，控制人物状态、场景变化与情节节点；\n- **15 项多模态参考**：单次任务里组合 10 张图片、5 段视频、7 个主体，让角色、动作、镜头和声音的呈现贴近预期；\n- **视频精准编辑**：对原视频的主体与背景做增加、修改和删除，还能调整风格、天气、颜色、材质、景别视角；\n- **2 分钟超长续拍**：多次续拍把故事延展至 2 分钟（官方标注即将上线）；\n- **创意复刻**：智能分析参考视频的镜头语言与叙事结构，生成适配社媒爆款、商品广告等场景的全新商业视频。\n\n这套组合拳的指向很明确：把创作者从「反复抽卡、挑一条能用的」推向「按叙事意图精确生产」。多语种多口音（含北京、粤语、东北、四川等方言口音）和动态画面中的多语种文字生成，则是在为商业化交付补齐细节。\n\n## 行业视角：上市前夜，追赶字节\n\n网易的报道点出了这场发布的时间点：可灵在潜在香港上市前夕推出 Kling 4.0，加速追赶字节跳动的 Seedance（[网易报道](https:\u002F\u002Fwww.163.com\u002Fdy\u002Farticle\u002FL7VEJGRM05198NMR.html)）。资本动作也在同一节奏上——据 IT之家，7 月初始投资者同意向北京可灵注资 20.28 亿美元，8 月底国家人工智能基金再注入 14 亿元现金资本。\n\n视频生成赛道此刻挤满了重量级玩家，而在单帧画质普遍够用之后，可控性、叙事长度与交付规格正在取代画质本身，成为新的竞争主线。Kling 4.0 把关键帧、多模态参考和 30 秒时长打包进同一代产品，抢的正是这条新主线的话语权。\n\n## 所以呢\n\n当生成时长跨过 30 秒、控制粒度细到关键帧，视频生成模型的对手已经不只是同行，而是传统制作流程本身。对创作者而言，值得问的问题也变了：不再是「AI 能不能生成视频」，而是「我的叙事，能不能拆成 10 个关键帧」。","https:\u002F\u002Fwww.ithome.com\u002F1\u002F008\u002F073.htm","d07c7afa-c9fa-47a3-879b-08a3dadfd498",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"eaf35b67-8c08-4d6a-8567-90ee14f1175d","kling",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"7e89b5cc-57db-4f37-bc6d-28919a73931c","model-release",{"id":19,"name":20,"slug":20,"description":14,"color":14},"b1853a5a-d940-42b7-94f9-0488ee3f2cf7","new-model",{"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},"43889cc0-fe47-42a7-ae62-f75a51da8cf6","en","Kling 4.0: 30-Second 4K Video in One Shot","Kuaishou's Kling 4.0 generates up to 30s native video, with 10 keyframes, 15 multimodal refs, 4K 10-bit HDR and 2-min extend. Flash live now.","The ceiling on AI video generation just moved: single-shot duration jumped from the industry's typical few seconds to 30 seconds of native output. Kuaishou's Kling opened Kling 4.0 Flash to Black Gold annual members on September 28, with the full version slated for October — and that 30-second figure is the hardest number in this release ([full IT Home report](https:\u002F\u002Fwww.ithome.com\u002F1\u002F008\u002F073.htm)).\n\n## 30 Seconds, Plus Broadcast-Grade Specs\n\nThe official announcement deserves a number-by-number read: single-generation videos up to 30 seconds with long takes and continuous narrative; output covering 4K and 1080p 10-bit HDR (listed as \"coming soon\") plus 21:9 widescreen; prompt input up to 8,000 tokens, with the model also expanding and polishing prompts. On audio, it supports high-quality dual-channel stereo and refined lip-sync matching — what the company calls a sound closer to on-set recording.\n\nFor professional creators, 10-bit HDR plus 21:9 means more than \"sharper\": the official claim is no blown highlights against the light, no mushy night scenes, no halos around neon — leaving more room for color grading. That is a signal of moving toward film-industry delivery standards, not just consumer-grade feature stuffing.\n\n## From Slot Machine to Director's Chair\n\nIf the last generation of video models competed on image quality, Kling 4.0 clearly leans into controllability. The official capability list reads like a director's console:\n\n- **Multi-keyframe control**: up to 10 keyframe images to set shot nodes, controlling character states, scene changes and plot points;\n- **15 multimodal references**: combine 10 images, 5 video clips and 7 subjects in a single generation task;\n- **Precise video editing**: add, modify or remove subjects and backgrounds; adjust style, weather, color, material and shot scale;\n- **2-minute extended takes**: chain continuations up to 2 minutes (listed as coming soon);\n- **Creative replication**: analyze a reference video's camera language and narrative structure, then generate new commercial videos for social-media hits, product ads and similar scenarios.\n\nThe direction is explicit: shift creators from \"reroll until one clip works\" to \"produce precisely against narrative intent.\" Multilingual output with regional accents (including Beijing, Cantonese, Dongbei and Sichuan) and stable multilingual text rendering in dynamic shots fill in the commercial-delivery details.\n\n## Industry View: Pre-IPO, Chasing ByteDance\n\nNetEase's report framed the timing: Kling launched 4.0 ahead of a potential Hong Kong listing, accelerating its chase of ByteDance's Seedance ([NetEase report](https:\u002F\u002Fwww.163.com\u002Fdy\u002Farticle\u002FL7VEJGRM05198NMR.html)). The capital moves follow the same rhythm — per IT Home, initial investors agreed in July to inject $2.028 billion into Beijing Kling, and the state AI fund added RMB 1.4 billion in late August.\n\nThe video-generation track is crowded with heavyweights, and with single-frame quality broadly sufficient, controllability, narrative length and delivery specs are replacing image quality as the new competitive axis. Kling 4.0 bundles keyframes, multimodal references and 30-second duration into one release — aiming squarely at the narrative rights to that new axis.\n\n## So What\n\nWhen generation length crosses 30 seconds and control granularity reaches the keyframe, AI video's competitor is no longer just other models but the traditional production pipeline itself. The question worth asking has changed too: not \"can AI generate video,\" but \"can my story be broken into 10 keyframes?\"","kling-4-0-30s-4k-video","2026-10-03T23:10:00Z","2026-10-03T23:10:07.724746Z","2026-10-03T23:10:07.724760Z",true,"agent",451,[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},"c463258a-094c-4217-8130-eef2bfacf78c","Xmax X2.0 把实时交互视频模型端侧化:逐帧自回归 + 消费级显卡跑 960p@24fps","xmax-x2-realtime-video","2026-07-16T08:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"43644653-55a5-40e6-b48c-dd9a548b7311","可灵 3.0 Turbo 落地：把视频生成拆成「快速预览 + 影院成片」两段式工作流","kling-3-0-turbo-two-stage-workflow","2026-06-22T00:04:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"36d4991a-f6ed-4edf-8f9a-e0fcb1212044","可灵团队提出 AnchorWorld：用 3D 人体运动重塑「第一人称世界模拟」","anchorworld-kling-3d-motion-first-person","2026-06-08T10:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"e8ad36b6-ec3c-4c90-8fbc-bf1e6807ae34","xAI Grok Imagine Video 1.5：单图生视频登顶 Arena榜首，自回归 MoE 改写视频生成规则","grok-imagine-1-5-ar-moe-arena-top","2026-06-07T08:00:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"bef7a773-36fb-406b-949d-8d728b97b242","Matrix-Game 4.0 开源:单卡 60FPS 的实时可玩游戏世界","matrix-game-4-0-open-source-60fps","2026-10-21T13:09:32+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"240d8a4c-c025-4080-be14-06099f650a71","xAI 把视频价格打到底:1080p 蒸馏版 Lite 比 Veo 3.1 便宜 65%","grok-imagine-1-5-lite-cheaper-than-veo","2026-10-10T08:00:00+00:00"]