[{"data":1,"prerenderedAt":729},["ShallowReactive",2],{"news-\u002Fnews\u002Fsuno-v3-ai-music":3,"news-prev-\u002Fnews\u002Fsuno-v3-ai-music":170,"news-next-\u002Fnews\u002Fsuno-v3-ai-music":237,"news-related-\u002Fnews\u002Fsuno-v3-ai-music":318},{"id":4,"title":5,"body":6,"category":155,"date":156,"description":157,"draft":158,"extension":159,"meta":160,"navigation":161,"path":162,"seo":163,"source":164,"stem":165,"tags":166,"__hash__":169},"news\u002Fnews\u002Fsuno-v3-ai-music.md","Suno v3 让 AI 音乐出圈:输入文字生成完整歌曲",{"type":7,"value":8,"toc":147},"minimark",[9,13,17,20,23,26,29,113,116,119,122,125,128],[10,11,12],"p",{},"Suno 在 2024 年 3 月推出 v3 模型,输入一段文字描述就能生成带人声的完整歌曲,质量接近 Demo 级别。AI 音乐创作从概念验证进入实用阶段,普通人不需要乐理也能做出一首能听的歌。",[14,15,16],"h2",{"id":16},"事件详情",[10,18,19],{},"Suno 是一家波士顿的 AI 音乐创业公司,产品形态很简单:输入一句\"给我写一首关于加班的摇滚歌\",30 秒后给你两首完整的歌——有人声、有伴奏、有副歌,每首大约 2 分钟。v3 是 2024 年 3 月发布的版本,相比 v2 在人声清晰度和音乐结构上有明显提升,中文歌词的处理也比之前顺畅。",[10,21,22],{},"v3 之前,Suno 只在 Discord 里运行。v3 发布后上线了独立网站和付费方案:免费用户每天 10 首歌(50 积分),Pro 计划 10 美元\u002F月(2500 积分,约 500 首),Premier 30 美元\u002F月(10000 积分)。商业用途需要订阅 Pro 以上版本,免费版生成的作品不能商用。",[10,24,25],{},"版权是绕不开的问题。Suno 不公开训练数据,但生成的作品风格能明显听出流行音乐的痕迹。2024 年 6 月,美国唱片业协会(RIAA)代表环球、华纳、索尼三大唱片公司起诉 Suno 和 Udio,指控他们未经授权使用受版权保护的录音制品训练模型。案件至今仍在审理。",[14,27,28],{"id":28},"关键信息",[30,31,32,45],"table",{},[33,34,35],"thead",{},[36,37,38,42],"tr",{},[39,40,41],"th",{},"项目",[39,43,44],{},"参数",[46,47,48,57,65,73,81,89,97,105],"tbody",{},[36,49,50,54],{},[51,52,53],"td",{},"模型版本",[51,55,56],{},"Suno v3",[36,58,59,62],{},[51,60,61],{},"发布时间",[51,63,64],{},"2024 年 3 月",[36,66,67,70],{},[51,68,69],{},"单次生成",[51,71,72],{},"2 首 × 2 分钟左右",[36,74,75,78],{},[51,76,77],{},"免费额度",[51,79,80],{},"每天 10 首(50 积分)",[36,82,83,86],{},[51,84,85],{},"Pro 价格",[51,87,88],{},"10 美元\u002F月(2500 积分)",[36,90,91,94],{},[51,92,93],{},"Premier 价格",[51,95,96],{},"30 美元\u002F月(10000 积分)",[36,98,99,102],{},[51,100,101],{},"商用要求",[51,103,104],{},"Pro 及以上订阅",[36,106,107,110],{},[51,108,109],{},"主要竞品",[51,111,112],{},"Udio、网易天音",[14,114,115],{"id":115},"我们的观察",[10,117,118],{},"我们试过用 Suno 给一个产品宣传视频配背景音乐,生成的成品直接能用,省了买版权音乐的钱。但要注意两点:一是人声歌词容易出怪腔怪调,中文歌的咬字偶尔会像外国人在唱;二是结构重复,你生成 10 首摇滚,前奏和副歌的走向会撞车,听多了能听出\"AI 味\"。",[10,120,121],{},"工具定位很关键。Suno 适合做草稿、做内部 demo、做不需要严肃版权的物料,不适合直接做发歌发片。一旦涉及商用,版权风险就得自己背——RIAA 的诉讼结果没出来之前,大公司不敢用 Suno 做正式音乐资产。",[10,123,124],{},"AI 音乐对独立音乐人是个双刃剑。一方面它能快速把灵感变成可听的 demo,不必先找编曲;另一方面低门槛生成会挤压低端外包市场,做 BGM、做短视频配乐的人最先受冲击。",[14,126,127],{"id":127},"相关工具",[129,130,131,140],"ul",{},[132,133,134,139],"li",{},[135,136,138],"a",{"href":137},"\u002Ftools\u002Faudio\u002Fsuno","Suno"," — AI 音乐生成工具",[132,141,142,146],{},[135,143,145],{"href":144},"\u002Ftools\u002Faudio\u002Felevenlabs","ElevenLabs"," — AI 语音合成,可做配音",{"title":148,"searchDepth":149,"depth":149,"links":150},"",2,[151,152,153,154],{"id":16,"depth":149,"text":16},{"id":28,"depth":149,"text":28},{"id":115,"depth":149,"text":115},{"id":127,"depth":149,"text":127},"产品发布","2024-03-22","Suno 2024年推出 v3 模型,输入文字描述就能生成带人声的完整歌曲,质量接近 Demo 级别,AI 音乐创作从概念验证进入实用阶段。",false,"md",{},true,"\u002Fnews\u002Fsuno-v3-ai-music",{"title":5,"description":157},"https:\u002F\u002Fsuno.com","news\u002Fsuno-v3-ai-music",[138,167,168],"AI音乐","创作工具","Hj5D6VoqFlIzUC89Azpj6uFtR3526SYf8Dlz1P8Epn0",{"id":171,"title":172,"body":173,"category":224,"date":225,"description":226,"draft":158,"extension":159,"meta":227,"navigation":161,"path":228,"seo":229,"source":195,"stem":230,"tags":231,"__hash__":236},"news\u002Fnews\u002Fweekly\u002F2024-03-17.md","AI 周报 · 2024-03-17 当周权威信源更新",{"type":7,"value":174,"toc":220},[175,178,182,198,201,205,217],[10,176,177],{},"本期汇总 2024-03-17 当周来自 OpenAI News 的 2 条更新。内容为信源标题与摘要的客观转述，完整信息请点原文链接。",[14,179,181],{"id":180},"_1-saving-lives-with-ai-health-coaching","1. Saving lives with AI health coaching",[129,183,184,187,190],{},[132,185,186],{},"来源：OpenAI News（原文日期 2024-03-13）",[132,188,189],{},"分类：市场动态",[132,191,192,193],{},"原文：",[135,194,195],{"href":195,"rel":196},"https:\u002F\u002Fopenai.com\u002Findex\u002Fhealthify",[197],"nofollow",[10,199,200],{},"Healthify collaborates with OpenAI to improve millions of lives with sustainable weight loss.",[14,202,204],{"id":203},"_2-global-news-partnerships-le-monde-and-prisa-media","2. Global news partnerships: Le Monde and Prisa Media",[129,206,207,209,211],{},[132,208,186],{},[132,210,189],{},[132,212,192,213],{},[135,214,215],{"href":215,"rel":216},"https:\u002F\u002Fopenai.com\u002Findex\u002Fglobal-news-partnerships-le-monde-and-prisa-media",[197],[10,218,219],{},"We have partnered with international news organizations Le Monde and Prisa Media to bring French and Spanish news content to ChatGPT.",{"title":148,"searchDepth":149,"depth":149,"links":221},[222,223],{"id":180,"depth":149,"text":181},{"id":203,"depth":149,"text":204},"AI 周报","2024-03-17","2024-03-17 当周 AI 行业权威信源更新汇总，共 2 条，来源：OpenAI News。",{},"\u002Fnews\u002Fweekly\u002F2024-03-17",{"title":172,"description":226},"news\u002Fweekly\u002F2024-03-17",[232,233,234,235],"AI周报","每周更新","权威信源","AI资讯","6yfNKzjUPuZLOMjW6i7zabOTKDWe_hwyy-B7BJl38_Y",{"id":238,"title":239,"body":240,"category":224,"date":310,"description":311,"draft":158,"extension":159,"meta":312,"navigation":161,"path":313,"seo":314,"source":261,"stem":315,"tags":316,"__hash__":317},"news\u002Fnews\u002Fweekly\u002F2024-03-24.md","AI 周报 · 2024-03-24 当周权威信源更新",{"type":7,"value":241,"toc":305},[242,245,249,263,266,270,283,286,290,302],[10,243,244],{},"本期汇总 2024-03-24 当周来自 OpenAI News 的 3 条更新。内容为信源标题与摘要的客观转述，完整信息请点原文链接。",[14,246,248],{"id":247},"_1-embedding-ai-into-developer-software","1. Embedding AI into developer software",[129,250,251,254,257],{},[132,252,253],{},"来源：OpenAI News（原文日期 2024-03-21）",[132,255,256],{},"分类：产品发布",[132,258,192,259],{},[135,260,261],{"href":261,"rel":262},"https:\u002F\u002Fopenai.com\u002Findex\u002Fjetbrains",[197],[10,264,265],{},"JetBrains uses OpenAI’s API to build its fastest-growing product ever.",[14,267,269],{"id":268},"_2-enterprise-ready-trust-and-safety","2. Enterprise-ready trust and safety",[129,271,272,275,277],{},[132,273,274],{},"来源：OpenAI News（原文日期 2024-03-18）",[132,276,256],{},[132,278,192,279],{},[135,280,281],{"href":281,"rel":282},"https:\u002F\u002Fopenai.com\u002Findex\u002Fsalesforce",[197],[10,284,285],{},"Salesforce integrates OpenAI’s enterprise-ready LLMs to transform customer applications.",[14,287,289],{"id":288},"_3-building-a-data-driven-efficient-culture-with-ai","3. Building a data-driven, efficient culture with AI",[129,291,292,294,296],{},[132,293,274],{},[132,295,256],{},[132,297,192,298],{},[135,299,300],{"href":300,"rel":301},"https:\u002F\u002Fopenai.com\u002Findex\u002Fholiday-extras",[197],[10,303,304],{},"Holiday Extras rolls out ChatGPT Enterprise across every team, boosting productivity by 500 hours weekly.",{"title":148,"searchDepth":149,"depth":149,"links":306},[307,308,309],{"id":247,"depth":149,"text":248},{"id":268,"depth":149,"text":269},{"id":288,"depth":149,"text":289},"2024-03-24","2024-03-24 当周 AI 行业权威信源更新汇总，共 3 条，来源：OpenAI News。",{},"\u002Fnews\u002Fweekly\u002F2024-03-24",{"title":239,"description":311},"news\u002Fweekly\u002F2024-03-24",[232,233,234,235],"r-RHG0gYTMo2dOEAnzVhuU85eusN8dzP2z8h6qw7_9E",[319,462,590],{"id":320,"title":321,"body":322,"category":155,"date":451,"description":452,"draft":158,"extension":159,"meta":453,"navigation":161,"path":454,"seo":455,"source":456,"stem":457,"tags":458,"__hash__":461},"news\u002Fnews\u002Fanthropic-claude-35-sonnet.md","Anthropic 发布 Claude 3.5 Sonnet,代码能力超过 GPT-4o",{"type":7,"value":323,"toc":445},[324,327,329,332,335,338,340,416,418,421,424,427,429],[10,325,326],{},"Anthropic 在 2024 年 6 月 20 日发布 Claude 3.5 Sonnet,在编码和推理基准测试上超过 GPT-4o,同时推出 Artifacts 功能,可在对话中实时生成和预览代码、网页、图表。该模型免费向 Claude.ai 用户开放。",[14,328,16],{"id":16},[10,330,331],{},"Claude 3.5 Sonnet 是 Anthropic Claude 3.5 系列的第一款模型。发布当天,它在 SWE-bench(软件工程基准测试,让 AI 修 GitHub 上的真实 bug)上拿到 31.2% 的成绩,超过 GPT-4o(约 22%)和自家上一代 Claude 3 Opus。GPQA(研究生级推理)和 MMLU(知识问答)等基准上同样排在第一梯队。",[10,333,334],{},"价格反而比上一代低。Claude 3.5 Sonnet 的 API 输入 3 美元\u002F百万 token,输出 15 美元\u002F百万 token,和 Claude 3 Sonnet 同价,但能力接近甚至超过更贵的 Claude 3 Opus。",[10,336,337],{},"Artifacts 是这次发布的另一个亮点。用户让 Claude 写一段 React 代码、一个 SVG 图表或一份 HTML 页面,模型会在对话窗口右侧生成一个独立面板,实时渲染结果,可以交互、可以复制。之前这类需求要来回复制代码到本地跑,现在对话里直接看效果。",[14,339,28],{"id":28},[30,341,342,350],{},[33,343,344],{},[36,345,346,348],{},[39,347,41],{},[39,349,44],{},[46,351,352,360,368,376,384,392,400,408],{},[36,353,354,357],{},[51,355,356],{},"模型名",[51,358,359],{},"Claude 3.5 Sonnet",[36,361,362,365],{},[51,363,364],{},"发布日期",[51,366,367],{},"2024 年 6 月 20 日",[36,369,370,373],{},[51,371,372],{},"SWE-bench 得分",[51,374,375],{},"31.2%(当时最高)",[36,377,378,381],{},[51,379,380],{},"API 输入价格",[51,382,383],{},"3 美元\u002F百万 token",[36,385,386,389],{},[51,387,388],{},"API 输出价格",[51,390,391],{},"15 美元\u002F百万 token",[36,393,394,397],{},[51,395,396],{},"新功能",[51,398,399],{},"Artifacts(对话内实时渲染)",[36,401,402,405],{},[51,403,404],{},"上下文窗口",[51,406,407],{},"200K token",[36,409,410,413],{},[51,411,412],{},"免费可用",[51,414,415],{},"是(Claude.ai 有用量限制)",[14,417,115],{"id":115},[10,419,420],{},"我们日常写代码主要用 Claude 3.5 Sonnet,几个对比结论:写前端组件它比 GPT-4o 强,尤其是 React + Tailwind 的组合,一次出的代码改两行就能跑;长上下文处理(200K)对读整个仓库很有用,GPT-4o 当时还是 128K。中文写作它不如国产模型接地气,但英文输出明显比 GPT-4o 干净,少翻译腔。",[10,422,423],{},"Artifacts 这个功能我们用了一段时间,体验比想象中好。改一个图表的配色,它边写边渲染,你能立刻看到效果。这个交互方式后来被不少产品借鉴,Cursor 的 Composer、v0 都有类似设计。",[10,425,426],{},"定价策略值得注意。Anthropic 把中端型号(Sonnet)卖到接近旗舰的能力,价格还压着 OpenAI,这是在抢开发者市场。Claude 3.5 Opus 和 Haiku 之前没出,2024 年 10 月才补上 3.5 Haiku,Opus 至今没出 3.5 版本。",[14,428,127],{"id":127},[129,430,431,438],{},[132,432,433,437],{},[135,434,436],{"href":435},"\u002Ftools\u002Fchat\u002Fclaude","Claude"," — Anthropic 的对话式 AI 助手",[132,439,440,444],{},[135,441,443],{"href":442},"\u002Ftools\u002Fcoding\u002Fcursor","Cursor"," — AI 代码编辑器,支持切换到 Claude 3.5 Sonnet",{"title":148,"searchDepth":149,"depth":149,"links":446},[447,448,449,450],{"id":16,"depth":149,"text":16},{"id":28,"depth":149,"text":28},{"id":115,"depth":149,"text":115},{"id":127,"depth":149,"text":127},"2024-06-20","Anthropic 2024年6月20日发布 Claude 3.5 Sonnet,在编码和推理基准测试上超过 GPT-4o,同时推出 Artifacts 功能,可在对话中实时生成和预览代码。",{},"\u002Fnews\u002Fanthropic-claude-35-sonnet",{"title":321,"description":452},"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-3-5-sonnet","news\u002Fanthropic-claude-35-sonnet",[459,436,460],"Anthropic","AI编程","YNpSEVyGXxoeLs4EYq9596u4EkM39BEPp7_8srryhco",{"id":463,"title":464,"body":465,"category":155,"date":578,"description":579,"draft":158,"extension":159,"meta":580,"navigation":161,"path":581,"seo":582,"source":583,"stem":584,"tags":585,"__hash__":589},"news\u002Fnews\u002Fopenai-gpt-4o-launch.md","OpenAI 发布 GPT-4o:实时语音对话+多模态,免费用户可用",{"type":7,"value":466,"toc":572},[467,470,472,475,478,481,483,553,555,558,561,563],[10,468,469],{},"OpenAI 在 2024 年 5 月 13 日的春季发布会上发布 GPT-4o 模型,支持实时语音对话、图片和视频理解,免费用户可用。这是 OpenAI 首次把旗舰模型免费开放,API 价格比 GPT-4 Turbo 降了一半。",[14,471,16],{"id":16},[10,473,474],{},"GPT-4o 的 \"o\" 代表 omni(全能)。它在单一模型里处理文本、音频和图像三种输入,不再像之前那样把语音对话拆成三段(语音转文字、模型处理文字、文字转语音)。端到端处理的好处是保留了语气、情绪和背景音,响应也更快——平均延迟 320 毫秒,最低 232 毫秒,接近人类对话的反应速度(人类对话平均约 200-300 毫秒)。",[10,476,477],{},"发布会上 OpenAI 演示了几段实时对话。模型能听出呼吸声和笑声,在对话里即兴唱歌,还能通过摄像头看用户画的方程图,实时讲解解题步骤。这些演示一度引发争议,因为部分画面被指经过剪辑,但实际能力基本属实。",[10,479,480],{},"免费用户可以使用 GPT-4o,但有用量限制,达到上限后会切换回更小的模型。Plus 用户(20 美元\u002F月)上限更高,Team 和 Pro 用户依次递增。API 端,GPT-4o 的输入价格 5 美元\u002F百万 token,输出 15 美元\u002F百万 token,比 GPT-4 Turbo(输入 10 美元、输出 30 美元)便宜一半,速度也快了一倍。",[14,482,28],{"id":28},[30,484,485,493],{},[33,486,487],{},[36,488,489,491],{},[39,490,41],{},[39,492,44],{},[46,494,495,502,509,517,525,532,539,545],{},[36,496,497,499],{},[51,498,356],{},[51,500,501],{},"GPT-4o(\"o\" = omni)",[36,503,504,506],{},[51,505,364],{},[51,507,508],{},"2024 年 5 月 13 日",[36,510,511,514],{},[51,512,513],{},"支持输入",[51,515,516],{},"文本、音频、图像、视频",[36,518,519,522],{},[51,520,521],{},"语音响应延迟",[51,523,524],{},"平均 320ms,最低 232ms",[36,526,527,529],{},[51,528,412],{},[51,530,531],{},"是(有用量限制)",[36,533,534,536],{},[51,535,380],{},[51,537,538],{},"5 美元\u002F百万 token",[36,540,541,543],{},[51,542,388],{},[51,544,391],{},[36,546,547,550],{},[51,548,549],{},"相比 GPT-4 Turbo",[51,551,552],{},"降价 50%,速度提升 2 倍",[14,554,115],{"id":115},[10,556,557],{},"我们当时第一时间用上了 GPT-4o 的语音模式。它最大的变化不是\"能对话\",而是\"能听见情绪\"。你叹气它知道你不耐烦,你笑它也跟着笑。这种交互密度是之前语音助手做不到的,但延迟实测比官方数据高,网络差的时候会卡顿。",[10,559,560],{},"免费开放这件事的意义被低估了。之前免费用户只能用 GPT-3.5,和 Plus 差一档,现在免费用户和付费用户用同一个模型,差别只在用量。这个动作直接把竞品的免费档位卡死了——Google 后来跟进把 Gemini 免费开放,Anthropic 也放宽了 Claude 的免费额度。",[14,562,127],{"id":127},[129,564,565],{},[132,566,567,571],{},[135,568,570],{"href":569},"\u002Ftools\u002Fchat\u002Fchatgpt","ChatGPT"," — OpenAI 的对话式 AI 助手,GPT-4o 的主要载体",{"title":148,"searchDepth":149,"depth":149,"links":573},[574,575,576,577],{"id":16,"depth":149,"text":16},{"id":28,"depth":149,"text":28},{"id":115,"depth":149,"text":115},{"id":127,"depth":149,"text":127},"2024-05-13","OpenAI 2024年5月13日发布 GPT-4o 模型,支持实时语音对话、图片和视频理解,免费用户可用。这是 OpenAI 首次将旗舰模型免费开放。",{},"\u002Fnews\u002Fopenai-gpt-4o-launch",{"title":464,"description":579},"https:\u002F\u002Fopenai.com\u002Findex\u002Fhello-gpt-4o\u002F","news\u002Fopenai-gpt-4o-launch",[586,587,588],"OpenAI","GPT-4o","多模态","iIO4NZVUQbQPfdqLW6R9fOE27swfdyB0SY3OZdYSTnc",{"id":591,"title":592,"body":593,"category":155,"date":719,"description":720,"draft":158,"extension":159,"meta":721,"navigation":161,"path":722,"seo":723,"source":724,"stem":725,"tags":726,"__hash__":728},"news\u002Fnews\u002Fopenai-sora-preview.md","OpenAI 展示 Sora:AI 生成 60 秒视频,物理模拟接近真实",{"type":7,"value":594,"toc":713},[595,598,600,603,606,609,611,678,680,683,686,689,691],[10,596,597],{},"OpenAI 在 2024 年 2 月 15 日展示 Sora 视频生成模型,能根据文字提示生成长达 60 秒的视频,物理模拟能力接近真实,引发行业震动。该模型当时未公开发布,仅向红队和部分创作者开放测试。",[14,599,16],{"id":16},[10,601,602],{},"Sora 是 OpenAI 的文生视频模型。一段 prompt 进去,出来最长 60 秒的视频,画面里的人物走动、物体遮挡、光影变化基本符合物理规律。发布会放出的几个 demo——东京街头漫步的女性、纸船在浪花里起伏、几只长毛象在雪地行走——质量明显高于同期 Runway Gen-2 和 Pika。",[10,604,605],{},"技术上 Sora 用了 Diffusion Transformer 架构。把视频拆成时空 patch(类似语言模型的 token),在 patch 上做扩散生成,统一处理图像和视频。这种方式让模型能处理不同时长、分辨率、宽高比的素材,生成时也能保持长视频的时间连贯性。OpenAI 没公开训练数据,但外界普遍推测用了大量合成数据和带版权的影视素材,这也埋下了后来的版权官司。",[10,607,608],{},"模型发布后没有立即开放。OpenAI 采取红队测试策略,先让安全研究人员和少数创作者试用,逐步公开。普通用户等到 2024 年 12 月才正式用上 Sora,且仅限 Plus 和 Pro 用户。这个过程里 Google Veo、快手可灵、Runway Gen-3 都相继发布,逼着 OpenAI 加快进度。",[14,610,28],{"id":28},[30,612,613,621],{},[33,614,615],{},[36,616,617,619],{},[39,618,41],{},[39,620,44],{},[46,622,623,630,638,646,654,662,670],{},[36,624,625,627],{},[51,626,356],{},[51,628,629],{},"Sora",[36,631,632,635],{},[51,633,634],{},"首次展示日期",[51,636,637],{},"2024 年 2 月 15 日",[36,639,640,643],{},[51,641,642],{},"最大生成时长",[51,644,645],{},"60 秒",[36,647,648,651],{},[51,649,650],{},"架构",[51,652,653],{},"Diffusion Transformer(时空 patch)",[36,655,656,659],{},[51,657,658],{},"公开发布",[51,660,661],{},"2024 年 12 月(Plus\u002FPro 用户)",[36,663,664,667],{},[51,665,666],{},"定价",[51,668,669],{},"Plus 用户有限量,Pro 200 美元\u002F月",[36,671,672,675],{},[51,673,674],{},"同期竞品",[51,676,677],{},"Runway Gen-3、Pika、可灵、Google Veo",[14,679,115],{"id":115},[10,681,682],{},"我们当时看了所有 demo,第一反应是\"这下视频行业的版图要变\"。但用上一段时间后判断更冷静了:60 秒是个噱头,真正能用的片段大多在 10 秒以内,长视频里仍会出现人物变形、物体凭空消失这类问题。物理模拟\"接近真实\"是营销话术,复杂场景(多人物互动、精细手部动作)还是会露馅。",[10,684,685],{},"定价和算力是更大的门槛。生成一段视频的成本远高于生成图片,Plus 用户的配额很少,真正做商业化内容得用 Pro。这个成本结构让 Sora 短期内很难替代商业拍摄,更多是用在概念验证和素材草图阶段。",[10,687,688],{},"国内可灵、即梦跟得很紧,可灵在人物动作一致性上已经不输 Sora,而且有国内合规优势。视频生成这个赛道,技术差距在缩小,真正的分水岭是版权和算力成本谁先解决。",[14,690,127],{"id":127},[129,692,693,699,706],{},[132,694,695,698],{},[135,696,629],{"href":697},"\u002Ftools\u002Fvideo\u002Fsora"," — OpenAI 的视频生成模型",[132,700,701,705],{},[135,702,704],{"href":703},"\u002Ftools\u002Fvideo\u002Frunway","Runway"," — 老牌 AI 视频工具",[132,707,708,712],{},[135,709,711],{"href":710},"\u002Ftools\u002Fvideo\u002Fkling","可灵"," — 快手出品的视频生成模型",{"title":148,"searchDepth":149,"depth":149,"links":714},[715,716,717,718],{"id":16,"depth":149,"text":16},{"id":28,"depth":149,"text":28},{"id":115,"depth":149,"text":115},{"id":127,"depth":149,"text":127},"2024-02-15","OpenAI 2024年2月15日展示 Sora 视频生成模型,能根据文字提示生成长达 60 秒的视频,物理模拟能力接近真实,引发行业震动。",{},"\u002Fnews\u002Fopenai-sora-preview",{"title":592,"description":720},"https:\u002F\u002Fopenai.com\u002Fsora","news\u002Fopenai-sora-preview",[586,629,727],"AI视频","tKTYmFxh7CKXsrw9pjY4eF8MLxW9MqJe_gXZ28Bm5TY",1794695479810]