谷歌发布 Gemini Omni 1.2 Flash，视频最长15秒且价格降低30%
谷歌发布 Gemini Omni 1.2 Flash，最长可生成15秒视频，720p生成价格降至每秒0.07美元，并已接入部分产品。
Loading…
bibibi.ccbibibi.ccLoading…
Connect scattered updates into a story you can understand.
谷歌发布 Gemini Omni 1.2 Flash，最长可生成15秒视频，720p生成价格降至每秒0.07美元，并已接入部分产品。
动察 Beating AI 援引《智能涌现》称，字节跳动视频生成模型Seedance海外收入占比已达50%，过去半年海外收入增长近5倍。报道还称腾讯、B站为其中国客户，实体行业客户占比超过四分之一。
Beating AI News reports that Pika released H3 Flex, improving generation speed and image quality through fine-tuning, distillation, and preference training, while making adaptation weights and inference code public.
ByteDance has launched Jimeng Short Drama Studio, integrating screenwriting, character design, storyboarding, video generation, and post-production, and supporting the remaking of completed short dramas into overseas versions.
Beating reported that ByteDance’s CapCut plans to integrate video generation and editing tools into ChatGPT, and this year November open a public beta on mobile, desktop, and web platforms. Users can upload footage, generate an initial draft, and continue editing in the chat interface.
Volcano Engine releases Seedance supporting 4K-based Seedance 2.5, with the first batch targeted at 20 film and television studios and directors worldwide; approximately 20 works have already been trialed.
PANews reported that Kling’s Kling 4.0 API has officially launched and will gradually become available to enterprise customers and developers worldwide. The model has also been integrated into multiple products and platforms.
According to IT reports by IT Home, Doubao’s head of public relations, Liu Xing, said that the data claiming “more than half of China’s computing power is used to produce escapist dramas” are unreliable; approximately 10%of ByteDance’s total computing power is used for video generation model training and inference.
Ramp’s software purchasing list shows that TypeSafe AI reached the top of the breakout-growth ranking with Jev; enterprise adoption grew by approximately 1 months after launch, by about 1 percentage points. Video-generation tools such as Runway and Higgsfield also entered the top ten for market-share growth.
Tian Keyu disclosed a world-model startup plan. The team has secured nearly 30000000 USD in funding and plans to train its model on 100000000 hours of video. The team plans to first release a public demonstration and release the full model in 2027.
Reka releases 19000000000 parameter model Rho-1, which can process text, images, and video and output robot actions within the same context; it remains a research preview.
Wangsu Science & Technology announced that it plans to use 300000000 to subscribe for Sand.ai's Cayman A+ round preferred shares; during the interim period, it will first provide Shanghai Sandai Technology with 300000000 in convertible bonds and obtain warrants.