Gemini Veo 3.1 Launches Multi-Image Reference, Synthesizes Three Elements into a Video in One Go

Google has rolled out the Veo 3.1 video model to Gemini Pro/Ultra subscription users today, introducing a new "Ingredients to Video" mode that supports uploading three reference images simultaneously, extracting character, scene, and style features, and fusing them into an 8-second 1080p video. The generated content includes an embedded SynthID invisible watermark. Users can input text prompts on the web or mobile app and generate videos with one click. The system maintains character consistency across frames and coherent lighting.

Google's demonstration shows that three different angle selfies + a cyber city background + an oil painting style image can output a short film titled "Impressionist Future Street Walk," with zero deformation in facial features and clothing. Veo 3.1 also outputs native ambient sounds, supports control of the first and last frames, and video extension features.

Google stated that the multi-image reference feature is now fully available, with generation quotas consistent with existing subscription plans, and no additional paid plans have been announced.

Databricks Co-founder Konwinski Warns: The US AI Research Advantage Is Being Lost

Databricks Co-founder Andy Konwinski warned that the US is yielding AI research leadership to China, which poses an existential threat to democracy. He pointed out that feedback from Berkeley and Stanford PhD students shows that about half of the notable new AI ideas in the past year have come from Chinese teams, a significant increase in proportion. Konwinski co-founded the venture capital firm Laude with his partner in 2024 and runs a nonprofit accelerator called Laud.

Turing Award Winner LeCun Exits Meta: Large Models Are a Dead End, the Future Lies in World Models

Turing Award winner Yann LeCun has diverged from Meta on the direction of AI. As Meta's Chief AI Scientist, he recently publicly criticized large language models as a dead end and advocated for research into world models. His leaving rumors have drawn attention; he once led the fundamental AI research department FAIR and was considered a key intellectual advisor at the company.

Nadella: Microsoft Does Not Chase the Low-Profit AI Computing Market, Emphasizes Long-Term Development

Microsoft CEO Nadella emphasized that the company's AI and cloud computing strategy differs from Oracle, as it does not pursue short-term benefits through low-profit services, but instead focuses on building a broad customer ecosystem. Oracle, on the other hand, plans to compete through low-profit hosted transactions by 2028.

Weibo Launches VibeThinker-1.5B, a Low-Cost AI Model Challenging Large Language Models

The Weibo AI department has launched the open-source large model VibeThinker-1.5B, which has 1.5 billion parameters. The model is optimized based on Alibaba's Qwen2.5-Math-1.5B and performs well in math and code tasks. It is now freely available on platforms such as Hugging Face, and it follows the MIT license, supporting commercial use.

Li Feifei's World Labs Launches Its First Commercial 3D World Model Marble, Supporting Multiple Inputs for Generation

The world model Marble, the first commercial product from World Labs co-founded by AI expert Li Feifei, supports the generation of editable 3D environments using text, images, videos, and 3D layouts. The product offers a freemium and paid subscription model, applicable to games, film, VR, and other fields. Less than a year after the company completed a $230 million funding round, it shows a strong pace of development.