[{"data":1,"prerenderedAt":750},["ShallowReactive",2],{"news-\u002Fnews\u002Fopenai-gpt-4o-launch":3,"news-prev-\u002Fnews\u002Fopenai-gpt-4o-launch":161,"news-next-\u002Fnews\u002Fopenai-gpt-4o-launch":251,"news-related-\u002Fnews\u002Fopenai-gpt-4o-launch":330},{"id":4,"title":5,"body":6,"category":145,"date":146,"description":147,"draft":148,"extension":149,"meta":150,"navigation":151,"path":152,"seo":153,"source":154,"stem":155,"tags":156,"__hash__":160},"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":8,"toc":137},"minimark",[9,13,17,20,23,26,29,113,116,119,122,125],[10,11,12],"p",{},"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。",[14,15,16],"h2",{"id":16},"事件详情",[10,18,19],{},"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)。",[10,21,22],{},"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。",[10,24,25],{},"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。",[14,27,28],{"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",{},"模型名",[51,55,56],{},"GPT-4o(\"o\" = omni)",[36,58,59,62],{},[51,60,61],{},"发布日期",[51,63,64],{},"2024 年 5 月 13 日",[36,66,67,70],{},[51,68,69],{},"支持输入",[51,71,72],{},"文本、音频、图像、视频",[36,74,75,78],{},[51,76,77],{},"语音响应延迟",[51,79,80],{},"平均 320ms,最低 232ms",[36,82,83,86],{},[51,84,85],{},"免费可用",[51,87,88],{},"是(有用量限制)",[36,90,91,94],{},[51,92,93],{},"API 输入价格",[51,95,96],{},"5 美元\u002F百万 token",[36,98,99,102],{},[51,100,101],{},"API 输出价格",[51,103,104],{},"15 美元\u002F百万 token",[36,106,107,110],{},[51,108,109],{},"相比 GPT-4 Turbo",[51,111,112],{},"降价 50%,速度提升 2 倍",[14,114,115],{"id":115},"我们的观察",[10,117,118],{},"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。",[10,120,121],{},"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。",[14,123,124],{"id":124},"相关工具",[126,127,128],"ul",{},[129,130,131,136],"li",{},[132,133,135],"a",{"href":134},"\u002Ftools\u002Fchat\u002Fchatgpt","ChatGPT"," — OpenAI 的对话式 AI 助手,GPT-4o 的主要载体",{"title":138,"searchDepth":139,"depth":139,"links":140},"",2,[141,142,143,144],{"id":16,"depth":139,"text":16},{"id":28,"depth":139,"text":28},{"id":115,"depth":139,"text":115},{"id":124,"depth":139,"text":124},"产品发布","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。",false,"md",{},true,"\u002Fnews\u002Fopenai-gpt-4o-launch",{"title":5,"description":147},"https:\u002F\u002Fopenai.com\u002Findex\u002Fhello-gpt-4o\u002F","news\u002Fopenai-gpt-4o-launch",[157,158,159],"OpenAI","GPT-4o","多模态","iIO4NZVUQbQPfdqLW6R9fOE27swfdyB0SY3OZdYSTnc",{"id":162,"title":163,"body":164,"category":238,"date":239,"description":240,"draft":148,"extension":149,"meta":241,"navigation":151,"path":242,"seo":243,"source":186,"stem":244,"tags":245,"__hash__":250},"news\u002Fnews\u002Fweekly\u002F2024-05-12.md","AI 周报 · 2024-05-12 当周权威信源更新",{"type":7,"value":165,"toc":233},[166,169,173,189,192,196,210,213,217,230],[10,167,168],{},"本期汇总 2024-05-12 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。",[14,170,172],{"id":171},"_1-introducing-the-model-spec","1. Introducing the Model Spec",[126,174,175,178,181],{},[129,176,177],{},"来源：OpenAI News（原文日期 2024-05-08）",[129,179,180],{},"分类：产品发布",[129,182,183,184],{},"原文：",[132,185,186],{"href":186,"rel":187},"https:\u002F\u002Fopenai.com\u002Findex\u002Fintroducing-the-model-spec",[188],"nofollow",[10,190,191],{},"原文标题如上，详情见来源链接。",[14,193,195],{"id":194},"_2-understanding-the-source-of-what-we-see-and-hear-online","2. Understanding the source of what we see and hear online",[126,197,198,201,204],{},[129,199,200],{},"来源：OpenAI News（原文日期 2024-05-07）",[129,202,203],{},"分类：技术更新",[129,205,183,206],{},[132,207,208],{"href":208,"rel":209},"https:\u002F\u002Fopenai.com\u002Findex\u002Funderstanding-the-source-of-what-we-see-and-hear-online",[188],[10,211,212],{},"Today we’re introducing new technology to help researchers identify content created by our tools and joining the Coalition for Content Provenance and Authenticity Steering Committee to promote industry standards.",[14,214,216],{"id":215},"_3-our-approach-to-data-and-ai","3. Our approach to data and AI",[126,218,219,221,224],{},[129,220,200],{},[129,222,223],{},"分类：市场动态",[129,225,183,226],{},[132,227,228],{"href":228,"rel":229},"https:\u002F\u002Fopenai.com\u002Findex\u002Fapproach-to-data-and-ai",[188],[10,231,232],{},"Just over a year after launching ChatGPT, AI is changing how we live, work and learn. It’s also raised important conversations about data in the age of AI. More on our approach, a new Media Manager for creators and conte…",{"title":138,"searchDepth":139,"depth":139,"links":234},[235,236,237],{"id":171,"depth":139,"text":172},{"id":194,"depth":139,"text":195},{"id":215,"depth":139,"text":216},"AI 周报","2024-05-12","2024-05-12 当周 AI 行业权威信源更新汇总，共 3 条，来源：OpenAI News。",{},"\u002Fnews\u002Fweekly\u002F2024-05-12",{"title":163,"description":240},"news\u002Fweekly\u002F2024-05-12",[246,247,248,249],"AI周报","每周更新","权威信源","AI资讯","lNjhAX36GPnjm4LYzZC-4DybGcNp_bmpG7cc7AhcjA8",{"id":252,"title":253,"body":254,"category":238,"date":322,"description":323,"draft":148,"extension":149,"meta":324,"navigation":151,"path":325,"seo":326,"source":274,"stem":327,"tags":328,"__hash__":329},"news\u002Fnews\u002Fweekly\u002F2024-05-19.md","AI 周报 · 2024-05-19 当周权威信源更新",{"type":7,"value":255,"toc":317},[256,259,263,276,279,283,295,298,302,314],[10,257,258],{},"本期汇总 2024-05-19 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。",[14,260,262],{"id":261},"_1-improvements-to-data-analysis-in-chatgpt","1. Improvements to data analysis in ChatGPT",[126,264,265,268,270],{},[129,266,267],{},"来源：OpenAI News（原文日期 2024-05-16）",[129,269,180],{},[129,271,183,272],{},[132,273,274],{"href":274,"rel":275},"https:\u002F\u002Fopenai.com\u002Findex\u002Fimprovements-to-data-analysis-in-chatgpt",[188],[10,277,278],{},"Improvements to data analysis in ChatGPT Interact with tables and charts and add files directly from Google Drive and Microsoft OneDrive.",[14,280,282],{"id":281},"_2-openai-and-reddit-partnership","2. OpenAI and Reddit Partnership",[126,284,285,287,289],{},[129,286,267],{},[129,288,223],{},[129,290,183,291],{},[132,292,293],{"href":293,"rel":294},"https:\u002F\u002Fopenai.com\u002Findex\u002Fopenai-and-reddit-partnership",[188],[10,296,297],{},"OpenAI and Reddit Partnership We’re bringing Reddit’s unique content to ChatGPT and our products.",[14,299,301],{"id":300},"_3-creating-an-ai-powered-magic-studio","3. Creating an AI-powered Magic Studio",[126,303,304,306,308],{},[129,305,267],{},[129,307,203],{},[129,309,183,310],{},[132,311,312],{"href":312,"rel":313},"https:\u002F\u002Fopenai.com\u002Findex\u002Fcanva",[188],[10,315,316],{},"Canva is a visual communication platform, enjoyed by more than 175 million people monthly to make presentations, videos, documents, websites, social media graphics and more. A majority of the world’s knowledge workers la…",{"title":138,"searchDepth":139,"depth":139,"links":318},[319,320,321],{"id":261,"depth":139,"text":262},{"id":281,"depth":139,"text":282},{"id":300,"depth":139,"text":301},"2024-05-19","2024-05-19 当周 AI 行业权威信源更新汇总，共 3 条，来源：OpenAI News。",{},"\u002Fnews\u002Fweekly\u002F2024-05-19",{"title":253,"description":323},"news\u002Fweekly\u002F2024-05-19",[246,247,248,249],"je_BznsA79JYzFT-ikCmNcyAQGZr1c5n_dDyx_SwBEQ",[331,468,611],{"id":332,"title":333,"body":334,"category":145,"date":457,"description":458,"draft":148,"extension":149,"meta":459,"navigation":151,"path":460,"seo":461,"source":462,"stem":463,"tags":464,"__hash__":467},"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":335,"toc":451},[336,339,341,344,347,350,352,422,424,427,430,433,435],[10,337,338],{},"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。",[14,340,16],{"id":16},[10,342,343],{},"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。",[10,345,346],{},"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。",[10,348,349],{},"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。",[14,351,28],{"id":28},[30,353,354,362],{},[33,355,356],{},[36,357,358,360],{},[39,359,41],{},[39,361,44],{},[46,363,364,371,378,386,393,399,407,415],{},[36,365,366,368],{},[51,367,53],{},[51,369,370],{},"Claude 3.5 Sonnet",[36,372,373,375],{},[51,374,61],{},[51,376,377],{},"2024 年 6 月 20 日",[36,379,380,383],{},[51,381,382],{},"SWE-bench 得分",[51,384,385],{},"31.2%(当时最高)",[36,387,388,390],{},[51,389,93],{},[51,391,392],{},"3 美元\u002F百万 token",[36,394,395,397],{},[51,396,101],{},[51,398,104],{},[36,400,401,404],{},[51,402,403],{},"新功能",[51,405,406],{},"Artifacts(对话内实时渲染)",[36,408,409,412],{},[51,410,411],{},"上下文窗口",[51,413,414],{},"200K token",[36,416,417,419],{},[51,418,85],{},[51,420,421],{},"是(Claude.ai 有用量限制)",[14,423,115],{"id":115},[10,425,426],{},"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。",[10,428,429],{},"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。",[10,431,432],{},"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。",[14,434,124],{"id":124},[126,436,437,444],{},[129,438,439,443],{},[132,440,442],{"href":441},"\u002Ftools\u002Fchat\u002Fclaude","Claude"," — Anthropic 的对话式 AI 助手",[129,445,446,450],{},[132,447,449],{"href":448},"\u002Ftools\u002Fcoding\u002Fcursor","Cursor"," — AI 代码编辑器,支持切换到 Claude 3.5 Sonnet",{"title":138,"searchDepth":139,"depth":139,"links":452},[453,454,455,456],{"id":16,"depth":139,"text":16},{"id":28,"depth":139,"text":28},{"id":115,"depth":139,"text":115},{"id":124,"depth":139,"text":124},"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":333,"description":458},"https:\u002F\u002Fwww.anthropic.com\u002Fnews\u002Fclaude-3-5-sonnet","news\u002Fanthropic-claude-35-sonnet",[465,442,466],"Anthropic","AI编程","YNpSEVyGXxoeLs4EYq9596u4EkM39BEPp7_8srryhco",{"id":469,"title":470,"body":471,"category":145,"date":600,"description":601,"draft":148,"extension":149,"meta":602,"navigation":151,"path":603,"seo":604,"source":605,"stem":606,"tags":607,"__hash__":610},"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":472,"toc":594},[473,476,478,481,484,487,489,565,567,570,573,576,578],[10,474,475],{},"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。",[14,477,16],{"id":16},[10,479,480],{},"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。",[10,482,483],{},"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。",[10,485,486],{},"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。",[14,488,28],{"id":28},[30,490,491,499],{},[33,492,493],{},[36,494,495,497],{},[39,496,41],{},[39,498,44],{},[46,500,501,509,517,525,533,541,549,557],{},[36,502,503,506],{},[51,504,505],{},"模型版本",[51,507,508],{},"Suno v3",[36,510,511,514],{},[51,512,513],{},"发布时间",[51,515,516],{},"2024 年 3 月",[36,518,519,522],{},[51,520,521],{},"单次生成",[51,523,524],{},"2 首 × 2 分钟左右",[36,526,527,530],{},[51,528,529],{},"免费额度",[51,531,532],{},"每天 10 首(50 积分)",[36,534,535,538],{},[51,536,537],{},"Pro 价格",[51,539,540],{},"10 美元\u002F月(2500 积分)",[36,542,543,546],{},[51,544,545],{},"Premier 价格",[51,547,548],{},"30 美元\u002F月(10000 积分)",[36,550,551,554],{},[51,552,553],{},"商用要求",[51,555,556],{},"Pro 及以上订阅",[36,558,559,562],{},[51,560,561],{},"主要竞品",[51,563,564],{},"Udio、网易天音",[14,566,115],{"id":115},[10,568,569],{},"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\"。",[10,571,572],{},"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。",[10,574,575],{},"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。",[14,577,124],{"id":124},[126,579,580,587],{},[129,581,582,586],{},[132,583,585],{"href":584},"\u002Ftools\u002Faudio\u002Fsuno","Suno"," — AI 音乐生成工具",[129,588,589,593],{},[132,590,592],{"href":591},"\u002Ftools\u002Faudio\u002Felevenlabs","ElevenLabs"," — AI 语音合成,可做配音",{"title":138,"searchDepth":139,"depth":139,"links":595},[596,597,598,599],{"id":16,"depth":139,"text":16},{"id":28,"depth":139,"text":28},{"id":115,"depth":139,"text":115},{"id":124,"depth":139,"text":124},"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。",{},"\u002Fnews\u002Fsuno-v3-ai-music",{"title":470,"description":601},"https:\u002F\u002Fsuno.com","news\u002Fsuno-v3-ai-music",[585,608,609],"AI音乐","创作工具","Hj5D6VoqFlIzUC89Azpj6uFtR3526SYf8Dlz1P8Epn0",{"id":612,"title":613,"body":614,"category":145,"date":740,"description":741,"draft":148,"extension":149,"meta":742,"navigation":151,"path":743,"seo":744,"source":745,"stem":746,"tags":747,"__hash__":749},"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":615,"toc":734},[616,619,621,624,627,630,632,699,701,704,707,710,712],[10,617,618],{},"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。",[14,620,16],{"id":16},[10,622,623],{},"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。",[10,625,626],{},"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。",[10,628,629],{},"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。",[14,631,28],{"id":28},[30,633,634,642],{},[33,635,636],{},[36,637,638,640],{},[39,639,41],{},[39,641,44],{},[46,643,644,651,659,667,675,683,691],{},[36,645,646,648],{},[51,647,53],{},[51,649,650],{},"Sora",[36,652,653,656],{},[51,654,655],{},"首次展示日期",[51,657,658],{},"2024 年 2 月 15 日",[36,660,661,664],{},[51,662,663],{},"最大生成时长",[51,665,666],{},"60 秒",[36,668,669,672],{},[51,670,671],{},"架构",[51,673,674],{},"Diffusion Transformer(时空 patch)",[36,676,677,680],{},[51,678,679],{},"公开发布",[51,681,682],{},"2024 年 12 月(Plus\u002FPro 用户)",[36,684,685,688],{},[51,686,687],{},"定价",[51,689,690],{},"Plus 用户有限量,Pro 200 美元\u002F月",[36,692,693,696],{},[51,694,695],{},"同期竞品",[51,697,698],{},"Runway Gen-3、Pika、可灵、Google Veo",[14,700,115],{"id":115},[10,702,703],{},"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。",[10,705,706],{},"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。",[10,708,709],{},"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。",[14,711,124],{"id":124},[126,713,714,720,727],{},[129,715,716,719],{},[132,717,650],{"href":718},"\u002Ftools\u002Fvideo\u002Fsora"," — OpenAI 的视频生成模型",[129,721,722,726],{},[132,723,725],{"href":724},"\u002Ftools\u002Fvideo\u002Frunway","Runway"," — 老牌 AI 视频工具",[129,728,729,733],{},[132,730,732],{"href":731},"\u002Ftools\u002Fvideo\u002Fkling","可灵"," — 快手出品的视频生成模型",{"title":138,"searchDepth":139,"depth":139,"links":735},[736,737,738,739],{"id":16,"depth":139,"text":16},{"id":28,"depth":139,"text":28},{"id":115,"depth":139,"text":115},{"id":124,"depth":139,"text":124},"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":613,"description":741},"https:\u002F\u002Fopenai.com\u002Fsora","news\u002Fopenai-sora-preview",[157,650,748],"AI视频","tKTYmFxh7CKXsrw9pjY4eF8MLxW9MqJe_gXZ28Bm5TY",1794695479580]