![[54d10780-d322-4d84-8d46-086cf19fcc18.mp3]]

字节跳动发布Seedance 1.0 Pro，AI视频生成能力登顶行业榜首

简报：

• 字节跳动旗下火山引擎正式推出最新AI视频生成模型Seedance 1.0 Pro，其在文生视频和图生视频任务的表现超越Google Veo 3及可灵2.1，位居Artificial Analysis榜单首位。
• 新模型采用时间因果变分自编码器、解耦时空扩散变换器和多阶段蒸馏技术，实现高效、高质量的视频生成，支持1080p输出且生成速度快，成本低至3.67元/5秒。
• Seedance 1.0 Pro突出优势包括出色的画面细节、指令理解及多镜头效果，面向内容创作、电商、营销、影视等多行业应用，已通过火山引擎API开放调用。
• 市场反响积极，测试者称其画质和速度“已超越或媲美”Veо 3，成为AI视频生成领域的新行业标杆。

相关链接：

• Seedance 1.0 Pro发布：AI视频生成新标杆
• 字节Seedance 1.0 Pro产品新闻
• 火山引擎AI视频模型发布会回顾
• AI视频生成行业榜单解读
• Seedance 1.0 Pro技术深度解析

ByteDance Unveils Seedance 1.0 Pro, AI Video Generation Capabilities Crowned Industry Leader 🎥

Brief:

• ByteDance's Volcano Engine officially launched its latest AI video generation model, Seedance 1.0 Pro. Its performance in text-to-video and image-to-video tasks surpasses Google Veo 3 and Keling 2.1, ranking first on the Artificial Analysis leaderboard. 🚀
• The new model employs a temporal causal variational autoencoder, decoupled spatio-temporal diffusion transformer, and multi-stage distillation techniques to achieve efficient, high-quality video generation. It supports 1080p output with fast generation speeds and a low cost of just 3.67 yuan per 5 seconds.
• Seedance 1.0 Pro's key advantages include exceptional visual detail, robust instruction comprehension, and multi-shot effects. It targets multi-industry applications including content creation, e-commerce, marketing, and film & television, and is now available for public access via Volcano Engine's API.
• Market response has been overwhelmingly positive, with testers stating its video quality and speed "surpass or rival" Veo 3, establishing it as a new industry benchmark in AI video generation. ✨

Related Links:

• Seedance 1.0 Pro Released: A New Benchmark for AI Video Generation
• ByteDance Seedance 1.0 Pro Product News
• Volcano Engine AI Video Model Launch Event Recap
• Interpreting the AI Video Generation Industry Rankings
• In-depth Technical Analysis of Seedance 1.0 Pro

英伟达携手多国推进AI基础设施，助力欧洲算力十倍提升

简报：

• 英伟达CEO黄仁勋在巴黎GTC Europe大会宣布，与法国、德国、意大利、亚美尼亚等多国政府及企业签署多项AI基础设施与软件合作协议，重点推动“主权AI”架构，支持本地数据中心与AI服务建设；
• 欧洲正筹建超20个AI超级工厂，英伟达平台如GB200 NVL72及Grace Blackwell芯片将作为核心引擎，未来两年欧洲AI算力预计增长十倍；
• 合作包括与法国Mistral共建AI云平台（部署18,000颗芯片），在德国打造装备10,000颗GPU的“工业云”，并与电信及云服务商协同推进大语言模型与AI应用落地；
• 英伟达已吸引150万开发者和近万家企业加入生态系统，欧盟也设立专项基金推进AI基础建设，助力欧洲在全球AI竞争中跃升。

相关链接：

• 英伟达扩大欧洲AI布局，签下多项政府企业合作协议
• 黄仁勋：未来两年欧洲AI算力将增十倍
• NVIDIA推动主权AI与AI超级工厂建设
• 欧洲多国与英伟达共推AI基础设施

NVIDIA Partners with Multiple Nations to Advance AI Infrastructure, Aiming for Tenfold Increase in European Computing Power 🚀

Brief:

• NVIDIA CEO Jensen Huang announced at GTC Europe in Paris that the company has signed multiple AI infrastructure and software cooperation agreements with governments and enterprises from various countries, including France, Germany, Italy, and Armenia. The key focus is on promoting a "sovereign AI" architecture to support the construction of local data centers and AI services.
• Europe is planning over 20 AI "superfactories" 🏭, with NVIDIA platforms like GB200 NVL72 and Grace Blackwell chips serving as core engines. European AI computing power is projected to increase tenfold in the next two years.
• Collaborations include co-building an AI cloud platform with France's Mistral (deploying 18,000 chips), creating an "industrial cloud" equipped with 10,000 GPUs in Germany, and working with telecom and cloud service providers to accelerate the implementation of large language models and AI applications.
• NVIDIA has attracted 1.5 million developers and nearly 10,000 enterprises to its ecosystem. The EU has also established special funds to advance AI infrastructure, helping Europe ascend in the global AI competition. 🇪🇺

Related Links:

• NVIDIA Expands European AI Footprint, Signs Multiple Government and Enterprise Cooperation Agreements
• Jensen Huang: European AI Computing Power to Increase Tenfold in Next Two Years
• NVIDIA Promotes Sovereign AI and AI Superfactory Construction
• Multiple European Countries and NVIDIA Jointly Promote AI Infrastructure

OpenAI与多国投资者洽谈高达400亿美元新一轮融资

简报：

• OpenAI正在与沙特阿拉伯公共投资基金、印度信实集团以及现有股东阿布达比投资集团MGX商讨高达400亿美元的融资，推动模型开发及基础设施项目“星际之门”。
• 软银为领投方，OpenAI还计划向Coatue和Founders Fund分别募集至少1亿美元，并预计到2027年再融资170亿美元。
• OpenAI首席执行官山姆·奥特曼积极与印度和阿联酋官方磋商，以促进低成本AI生态建设并推进相关投资。

相关链接：

• OpenAI与沙特、印度投资者商讨巨额融资
• OpenAI寻求400亿美元融资以加速AI模型开发
• OpenAI加强布局未来，融资与技术双线推进
• OpenAI开源模型进展

OpenAI in Talks with International Investors for New Funding Round Up To $40 Billion 💰

Briefing:

• OpenAI is reportedly in discussions with Saudi Arabia's Public Investment Fund (PIF), India's Reliance Industries, and existing shareholder Abu Dhabi Investment Group MGX for a funding round of up to $40 billion. This substantial investment aims to propel model development and the ambitious "Stargate" infrastructure project. 🚀
• SoftBank is poised to be a lead investor, while OpenAI also plans to raise at least $100 million each from Coatue and Founders Fund. The company anticipates raising an additional$17 billion by 2027.
• OpenAI CEO Sam Altman has been actively consulting with officials in India and the UAE to promote the development of a low-cost AI ecosystem and drive related investments. 💡

Related Links:

• OpenAI Discussing Huge Funding with Saudi, Indian Investors
• OpenAI Seeks $40 Billion Funding to Accelerate AI Model Development
• OpenAI Strengthens Future Plans, Advancing Both Funding and Technology
• OpenAI Open-Source Model Progress

Meta发布V-JEPA2，开创视频AI理解与零样本机器控制新纪元

简报：

• 2025年6月11日，Meta正式发布全新视频理解模型V-JEPA2，由首席AI科学家Yann LeCun团队研发，具备自我监督学习和强大的零样本机器人控制能力。
• V-JEPA2通过超过100万小时的未标注视频训练，能够模拟人类“世界模型”认知，理解并预测物体的动态和相互关系，并可直接用于机器人任务，无需额外训练。
• 技术创新点包括自我监督、遮挡预测、抽象表征和高效迁移，使模型在视频内容分析、物理推理、动作识别等多项基准任务上表现远超传统方案。
• Meta已将V-JEPA2以开源方式发布，支持在全球不同平台使用，并配套提供三项标准化物理推理测试，推动视频AI和机器人领域研究进展。
• 官方计划未来扩展V-JEPA2长视频与音频分析能力，为AR眼镜和虚拟助手等智能设备提供支持。

相关链接：

• Meta's V-JEPA 2 model teaches AI to understand its surroundings
• Meta V-JEPA 2 world model uses raw video to train robots
• Meta Releases V-JEPA 2 AI World Model to Teach Robots Physics and Common Sense

Meta Releases V-JEPA2, Ushering in a New Era for Video AI Understanding and Zero-Shot Robot Control 🚀

Briefing:

• On June 11, 2025, Meta officially released its new video understanding model, V-JEPA2. Developed by Chief AI Scientist Yann LeCun's team, it features self-supervised learning and powerful zero-shot robot control capabilities. 🤖
• Trained on over 1 million hours of unlabeled video, V-JEPA2 can simulate human 'world model' cognition, understanding and predicting object dynamics and interrelationships, and can be directly applied to robotic tasks without additional training.
• Key technological innovations include self-supervision, occlusion prediction, abstract representation, and efficient transfer. These enable the model to significantly outperform traditional methods in various benchmark tasks such as video content analysis, physical reasoning, and action recognition. 💡
• Meta has open-sourced V-JEPA2, supporting its use on various global platforms. It also provides three standardized physical reasoning tests to advance research in video AI and robotics.
• Official plans include expanding V-JEPA2's long-form video and audio analysis capabilities in the future, providing support for smart devices like AR glasses and virtual assistants.

Related Links:

• Meta's V-JEPA 2 model teaches AI to understand its surroundings
• Meta V-JEPA 2 world model uses raw video to train robots
• Meta Releases V-JEPA 2 AI World Model to Teach Robots Physics and Common Sense

阿里巴巴开源3D数字人实时面捕框架Mnn3dAvatar，加速虚拟人应用落地

简报：

• 阿里巴巴近日正式开源3D数字人框架Mnn3dAvatar，为开发者提供面向直播、娱乐和商业场景的实时面部捕捉与3D虚拟角色生成能力。
• Mnn3dAvatar依托MNN深度学习推理框架，具备高效、轻量和多平台运行等特性，能够低延迟地同步用户表情至3D虚拟角色。
• 该框架已在淘宝、优酷等阿里系应用中验证稳定性，适用于直播带货、虚拟展示、教育培训等多种行业场景。
• 开源生态完备，支持安卓、iOS等平台，显著降低3D内容创作门槛，为内容创作者和企业提供支持。
• 项目详情与代码已开放给开发者社区。

相关链接：

• 项目详情及下载地址
• 新闻报道1

Alibaba Open-Sources Mnn3dAvatar: A Real-time 3D Digital Human Face Capture Framework to Accelerate Virtual Human Applications 🚀

Brief:

• Alibaba recently officially open-sourced its 3D digital human framework, Mnn3dAvatar, providing developers with real-time facial capture and 3D virtual character generation capabilities for live streaming, entertainment, and commercial scenarios. ✨
• Leveraging the MNN deep learning inference framework, Mnn3dAvatar boasts efficiency, lightweight design, and multi-platform compatibility, enabling low-latency synchronization of user expressions to 3D virtual characters.
• The framework's stability has been verified in Alibaba-affiliated applications such as Taobao and Youku, making it suitable for various industry scenarios, including live e-commerce, virtual displays, and education and training.
• With a complete open-source ecosystem, it supports platforms like Android and iOS, significantly lowering the barrier to 3D content creation and empowering content creators and businesses. 💡
• Project details and code are now accessible to the developer community.

Related Links:

• Project Details and Download Link
• News Report 1