内容来源：https://aibusiness.com/generative-ai/openai-intros-sora-2-social-media-app

内容总结：

近日，人工智能研究机构OpenAI正式发布新一代音视频生成模型Sora 2，并同步推出搭载该模型的社会化媒体应用Sora。此次升级着重提升了生成内容的真实感与可控性，试图解决前代产品存在的"失真感"问题。

据悉，Sora 2在画面质感、指令遵循及音效合成方面均有显著改进，支持生成电影级画面与动漫风格视频。配套推出的Sora应用现已在美国和加拿大地区上线iOS版本，用户可通过可定制化信息流进行视频创作、混剪及分享。基础版Sora 2将免费开放，但计算资源限制可能影响功能完整性，专业用户可通过订阅服务获取增强版体验。

行业分析师指出，此举体现OpenAI强化消费者生态的战略转向。Futurum集团分析师布拉德利·希明表示："通过免费策略与新应用组合，OpenAI正推动技术普及化。"不过专家也提醒，该技术仍面临多重挑战：Forrester分析师威廉·麦基翁-怀特指出实际工作流优化效果尚待验证；Gartner分析师阿伦·钱德拉塞卡则强调需关注深度伪造、版权争议等伦理风险。

值得注意的是，该应用在数据隐私保护方面已引发关注。OpenAI声明已针对青少年用户设置内容浏览限额，并强化"数字形象"功能权限管理。企业用户预计将延续通过微软Azure等可信平台接入模型的合规路径。

随着音视频生成技术加速融入社交生态，这场由Sora引发的创新浪潮将在创造新机遇的同时，对技术伦理与行业规范提出更严峻的考验。

中文翻译：

本文由谷歌云赞助
选择首个生成式AI应用场景
想要入门生成式AI，首先应关注能够优化人类信息交互体验的领域。

升级版多模态模型通过解决现实失真等问题提升真实感。该应用配备可自定义信息流，支持视频发现与混剪创作。
周二，OpenAI发布了其视频音频生成模型的最新版本Sora 2。这家生成式AI模型提供商还推出了由Sora 2驱动的新型社交媒体应用Sora。
距离初代音视频模型发布已逾一年，OpenAI此次推出的革新模型旨在解决困扰前代产品的技术难题。开发商指出，关键挑战在于避免过度美化或扭曲现实。
据OpenAI介绍，Sora 2显著提升了真实感，能够精准执行指令，擅长创作具有电影质感和动漫风格的逼真视频。该模型可生成真实的背景音景、语音及音效，并支持用户融入现实世界元素。
为方便用户体验，OpenAI同步推出了由Sora 2驱动的iOS版社交媒体应用Sora。通过该应用，用户可进行视频创作、相互混剪作品，并在个性化信息流中发现新鲜内容。目前该应用已在美国和加拿大上线。OpenAI表示Sora 2将免费开放，但免费版本可能因算力限制无法使用全部功能。ChatGPT Pro用户可在sora.com网站及即将更新的Sora应用中体验Sora 2 Pro版本。

市场研究机构Futurum Group分析师布拉德利·希明指出：“Sora体现了OpenAI的战略重点——增强用户参与度。通过新应用及免费策略，OpenAI正试图让该模型触达更广泛群体。”他强调，新版模型在生成逼真影像与声音方面取得显著突破，“这解决了AI长期存在的痛点，特别是视频生成中违和感强烈的问题。相较初代产品失真的表现，Sora 2有效摆脱了‘恐怖谷效应’，呈现出更自然的视觉效果。”

弗雷斯特研究公司分析师威廉·麦基翁-怀特认为，尽管模型在可控性、一致性和安全性方面的改进值得关注，但其实际应用效果仍需验证。“目前视频生成工作流需要反复调整提示词以接近创作预期，这个过程耗费惊人精力。若Sora 2能切实减少试错成本，将具有颠覆性意义。”

高德纳咨询公司分析师阿伦·钱德拉塞卡兰对比指出，相较于谷歌Veo视频生成模型，Sora 2更侧重速度与社交应用：“Veo凭借其稳定性和音画同步能力已成为成熟模型。”麦基翁-怀特补充道，Sora应用契合OpenAI产品服务多元化战略，将为模型训练提供优质数据集。但其成功与否取决于Sora 2能否在Reddit、YouTube和Instagram等平台竞争中快速获客。“虽然OpenAI忠实用户会率先采纳，但该公司初期成功很大程度上得益于Dall-E Mini引发的全民好奇风潮。”他提及这款已更名为Craiyon的图像生成应用时表示。

隐私安全专家同时发出警示，该应用在获得授权后将直接获取用户数据。希明提醒：“注重隐私安全的用户需谨慎使用此类应用。”企业用户可能沿用ChatGPT的部署模式，通过微软Azure或其他企业级可信平台接入模型。

随着社交媒体日益融入社会生态，希明预测Sora 2及其应用将同时吸引个人与企业用户，在创造机遇的同时也带来潜在风险。OpenAI声称已意识到相关责任，对青少年用户设置了每日生成内容浏览上限，并严格限制数字形象功能权限。通过数字形象功能，用户可自主管理形象授权，随时撤销使用许可或删除相关视频。

钱德拉塞卡兰最后强调，该技术仍面临伪造内容滥用、版权争议与知识产权冲突等挑战。

您可能还喜欢

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The updated multi-modal model aims to improve realism by addressing problems such as the distortion of reality. The app has a customizable feed for discovering and remixing videos.
OpenAI on Tuesday released Sora 2, the latest version of its video and audio generation model. The generative AI model provider also introduced Sora, a new social media app that is powered by Sora 2.
Sora 2 was released more than a year after OpenAI released the first generation of the audio and video model.
With the revamped model, OpenAI said it's seeking to address some of the problems that hampered previous video and audio models. One challenge is not being too optimistic or distorting reality, the vendor said.
Sora 2 is more realistic. It can follow instructions and is skilled at creating realistic, cinematic and anime-style videos, OpenAI said. It can create realistic background soundscapes, speech and sound effects, OpenAI said. Users can also inject elements of the real world.
To help users experiment with the model, OpenAI also launched a social media iOS app called Sora, which is powered by Sora 2.
With the app, users can create, remix each other's creations and discover new videos in a customizable feed. The iOS app is now available for download in the U.S. and Canada. OpenAI said Sora 2 will be available for free, but compute restrictions might limit the capabilities of the free version. ChatGPT Pro users will be able to use Sora 2 Pro on sora.com and soon in the Sora app.
"Sora emphasizes what I see as a priority for OpenAI, and that is engagement with consumers," said Bradley Shimmin, an analyst at the Futurum Group. He added that with the new app and making Sora 2 free for users, OpenAI is trying to get the model into as many hands as possible.
It is also clear that OpenAI has made the model more capable of producing realistic images and sounds, Shimmin said.
"It's reflective of something that AI has struggled with, particularly for video generation, which is an otherworldliness that doesn't quite seem right," he said, referring to the unrealistic qualities of Sora 1. "With this version of Sora, it looks less like an uncanny valley, and more like a happy plateau."
While improvements to controllability, consistency and safety are noteworthy, it will be interesting to evaluate the model in real-world applications, said William McKeon-White, an analyst at Forrester.
"So far, most workflows with video-gen have been about tuning prompts to get 'close enough' to the director or generators' vision, which has proven surprisingly labor-intensive," McKeon-White said. "If Sora 2 legitimately does reduce the back-and-forth required, that would be a game-changer."
Compared with Google Veo, another video generating model, Sora 2 focuses on speed and social use, said Gartner analyst Arun Chandrasekaran.
"Veo is certainly a much proven model at this point due to its consistency and audio-visual alignment," Chandrasekaran said.
The Sora app aligns well with OpenAI's strategy to diversify its product and services portfolio and will serve as an excellent training data set for OpenAI's development, McKeon-White said. However, he added that Sora's success will depend on Sora 2's ability to become popular quickly as it competes with other social media apps, such as Reddit, YouTube, and Instagram.
"There will be a loyal OpenAI contingent who will adopt it regardless, but a large part of OpenAI's preliminary success can be traced back to the sudden curiosity around, and popularity of, Dall-E Mini," McKeon-White said, referring to the app version of the vendor's first image-generating model, now rebranded as Craiyon.
The app also raises privacy and security concerns, as it grants OpenAI direct access to consumers' data if allowed.
"Consumers need to pay special attention to using applications like these if they are concerned about their privacy and security," Shimmin said.
Companies that employ this tool will likely take the same approach they used with ChatGPT, by accessing the model either using Microsoft Azure or another enterprise-grade and trusted medium.
However, because social media has become an integral part of consumers' and businesses' lives, Shimmin said Sora 2 and Sora will likely be used by both, creating opportunities and perils for users and those developing the technology.
OpenAI appears to be aware of the responsibility it holds and is placing default limits on the number of generations (OpenAI's term for media produced with Sora 2 and Sora) that teenagers can view per day, as well as stricter permissions on Cameos for teens, the vendor stated. With Cameos, users are in control of their likeness and can decide who can use it, as well as revoke access or remove any video featuring their Cameo at any time.
Other challenges include misuse such as the potential to create deepfakes, and copyright and IP conflict, Chandrasekaran said.
You May Also Like