AI Singularity.com丨AI News丨AI Tool Recommendation丨AI Tool Tutorial
Provide 1000+ AI efficiency tools丨 AIEVA.CN 丨Welcome to visit
AI Singularity reported on September 20, 2024丨AI Information Morning Post
Kuaishou's Keling AI released V1.5 version of the model: added "motion brush"
On September 19, Kuaishou's AI video creation platform "Kelin AI" released the Keling V1.5 model, adding the "motion brush" function. The Colin-V1.5 model is the first iteration of Colin's larger-digit version, which enhances the post-generation editing capabilities of AI videos in response to user feedback.
There are two major changes in this version upgrade. The first is that the image quality has been significantly improved, and it can output videos with a resolution of 1080p, and the second is the addition of the "Motion Brush" function, which was first created by the foreign AI video platform Runway and is favored by more and more users.
According to reports, the "motion brush" function specifies the motion trajectory for the elements in the picture (people or objects, etc.), and also supports the additional designation of static areas to improve the controllability of the picture.
The video quality of Coline AI has been improved, and at the same time, the composition is more beautiful, the action is more reasonable, the text relevance is improved, and it can also output a variety of mainstream video sizes.
At the same time as the release of this new version, the ColinAI team also announced other noteworthy new features that have been launched recently, including:
- Generate 4 short videos at once
- 10 seconds in length, support for adding tail frames
- Video quality enhancement is supported
Gai Kun, senior vice president of Kuaishou, disclosed at Kuaishou's investor day on September 13 a few days ago that so far, more than 2.6 million people have used Kuaishou's video generation model Keling AI, and more than 27 million videos have been generated. Since the product launch, Keline AI has undergone 9 iterations and upgrades.
Alibaba Cloud Tongyi Qianwen released the open source large model Qwen 2.5, and the flagship large model Qwen-Max was upgraded
On September 19, Alibaba Cloud held the annual Apsara Conference, at which Alibaba Cloud brought updates to a number of AI models and products of its Qwen series.
Qwen2.5, a new generation of open-source model of Tongyi Qianwen, was officially released, and the flagship model Qwen2.5-72B is claimed to surpass Llama 405B in performance.
The Qwen 2.5 series covers large language models, multimodal models, mathematical models, and code models of multiple sizes, and each model has a basic version, an instruction-following version, and a quantized version, with a total of more than 100 models on the shelves.
It is worth mentioning that at that time, Tongyi Qianwen's flagship large model Qwen-Max also ushered in an upgrade, claiming that its performance was close to OpenAI's GPT-4o. According to the official introduction, compared with the previous generation, Qwen-Max uses more training data, larger model size, and stronger human alignment instructions in training, and finally reaches a higher level of intelligence, with mathematical and code capabilities surpassing the GPT-4o version that was first released in May this year.
Alibaba Cloud's "Tongyi Wanxiang" launched the AI Wensheng video function
On September 19, at the Alibaba Cloud Apsara Conference held in Hangzhou, Alibaba Cloud's Tongyi Wanxiang visual content creation platform can generate a high-definition, realistic video in just a few minutes. This is the first time that Zhou Jingren, CTO of Alibaba Cloud, has demonstrated the powerful motion generation and concept combination capabilities of the Tongyi Wanxiang video generation model.
It is reported that Tongyi Wanxiang has also officially launched a video generation model, which is based on a Sora-like DiT architecture, which has been recognized by domestic AI companies.
Tongyi Wanxiang has opened up the two creative portals of "Wensheng Video" and "Tusheng Video", which support motion generation, complex semantic understanding and concept combination generation, as well as outline animation, Guoman 3D, felt animation, CG impasto, 3D animation, Chinese style painting style and elements and other styles.
According to reports, Tongyi Wanxiang's new video generation model can generate film-level high-definition videos, which can be applied to film and television creation, animation design, advertising design and other fields. From now on, all users can experience it for free through the Tongyi APP and Tongyi Wanxiang official website.
Wu Yongming, CEO of Alibaba: The imagination of AI can be extended to change the physical world
At Apsara 2024, Yongming Wu, CEO of Alibaba Group and Chairman and CEO of Alibaba Cloud Intelligence Group, gave a speech about his views on the future development of the AI industry.
Wu Yongming said that the greatest imagination of AI is not in the mobile phone screen, but by infiltrating, taking over the digital world, and changing the physical world, most things in the physical world will have AI capabilities, form the next generation of new products, and create synergies with the cloud AI-driven digital world.
He also believes that, like cars, there will be two or three robots in every urban home in the future to help people improve efficiency in their lives.
Generative AI consumes a lot of power: it takes 3 bottles of water to generate 100 words with GPT-4
Recently, according to the Washington Post, a new study from the University of California, Riverside in United States has revealed the staggering environmental costs of generative AI, especially its water consumption.
Studies have found that AI needs a lot of water to cool down servers, even just by generating text. This, combined with the enormous pressure it places on the power grid, further highlights the environmental challenges of AI development.
The study notes that water usage for AI varies by state and data center location, but overall, water consumption is inversely proportional to electricity bills. Texas has the lowest water consumption ratio, requiring only 235 milliliters of water to generate a 100-word email, while Washington has a whopping 1.408 milliliters, equivalent to three 16.9-ounce bottles of mineral water.
In addition, data centers themselves are large consumers of water and electricity, which also drives up the utility bills for residents in their areas. The cost of electricity for GPT-4 is also very high, and if one in 10 United States office workers use GPT-4 once a week, the electricity consumption for generative AI operations in one year would be equivalent to 20 days of electricity used by an entire Washington, D.C. resident.
[This article is from丨AI Singularity Network丨The same name of the whole network account丨Welcome to search and follow]
AI Singularity丨Provide 1000+ AI efficiency tools丨 AIEVA.CN 丨Welcome to visit