HomeAI News
AI highlights of the week: SD and Midjourney launched new versions

AI highlights of the week: SD and Midjourney launched new versions

Hayo News
Hayo News
June 25th, 2023

In the past week, Chinese friends had a happy Dragon Boat Festival holiday. The two "old friends" Stability AI and Midjourney have also launched new versions in the past few days, which can be regarded as bringing you something worth talking about. "Dragon Boat Festival gift".

OpenAI is exposed to launch 3 new features

Foreign media disclosed plans related to "AI App Store"

Although ChatGPT did not have much action this week, the news of OpenAI still did not stop.

Among them, a hacker dug deep into the source code of ChatGPT's client, and found 3 new functions : personal profile (let ChatGPT know you and remember you) , file upload, and workspace switching (let AI interact with different styles in different scenarios) , and according to the code, these new features may be available to us soon.

In addition, foreign media obtained more information about the previous OpenAI plan to establish a ChatGPT application platform similar to the "APP Store" . According to insiders, this "AI model store" will provide enterprises with access to the latest LLM, and enterprises can quickly find the most suitable large-scale AI models according to their needs.

Many people may think of the ChatGPT plug-in system launched earlier, but according to foreign media, the popularity of the plug-in system has not met expectations, and the plan for the "AI model store" may be full of challenges in the future.

Midjourney v5.2 released

The "Zoom Out" function shows its magic

Midjourney launched version 5.2 this week. In addition to the enhancement of basic skills such as the improvement of aesthetic style and the enhancement of text understanding, the newly launched "Zoom Out" function has attracted the attention of many people.

This function can simulate different focal lengths of the camera lens to "extend the drawing". Users can choose between three focal lengths: 28-35mm (0.5x zoom), 75mm (1.5x zoom) and 100mm (2x zoom). Small. There are also users who have used this new feature to create amazing short videos, which makes people admire the creative potential of "Zoom Out".

In addition, the "Make Square" (into a square) for creative cropping of pictures and the long-awaited prompt word analyzer "shorten command" are also online in version 5.2. I believe that the experience of Midjourney users will be greatly improved.

Stable Diffusion XL released version 0.9

The official version 1.0 is expected to be fully released in July

Stability AI announced the release of SDXL version 0.9 this week. Compared with the Beta version launched in April, version 0.9 has made significant progress in the quality of generated images and composition details.

(SDXL Beta on the left, SDXL 0.9 on the right)

Although the quality of the generated pictures is constantly improving, the hardware of SD XL has not improved much. The official hardware requirements of at least 16GB of memory and 8GB of 20-series N cards will surely allow most users to be able to use it when the official version is fully released in July. Give it a try.

GPT-Engineer open source release

Can automatically build the code base on demand

This week, a "dark horse" with 31.9k stars appeared on Github - the AI-driven code generation artifact GPT-Engineer, which can learn to imitate the user's programming style according to the user's needs, and automatically generate code, to build a complete code base.

And its most unique feature is that unlike other AIs that only accept the needs of users unilaterally, it will use GPT-4 to understand the ambiguities in the needs, and then ask detailed questions to allow users to clarify and detail. needs, and finally enters the construction of the code. To some extent, it seems that it is not just a "programmer", but also some of the work of a "product manager".

However, the current GPT-Engineer is not very stable in the actual measurement performance, and the process is a bit cumbersome, but as an emerging project, the follow-up development is worth looking forward to.

Zhang Yong steps down as Alibaba Group CEO

Focus on Alibaba Cloud, optimistic about the development of AI

"Alibaba" WeChat news, Zhang Yong announced that he will step down as the chairman and CEO of Alibaba Holding Group on September 10, and then he will serve as the chairman and CEO of Alibaba Cloud Intelligence Group full-time, focusing on the development of Alibaba Cloud. Cai Chongxin, currently the executive vice chairman of Alibaba Group, will serve as the chairman of the board of directors of Alibaba Holding Group.

Zhang Yong also mentioned in the letter to all employees that the complete spin-off of Alibaba Cloud Intelligence Group has started and is in the most critical period of upward development. At the "Alibaba Cloud Summit" in April this year, he announced in a high-profile manner that all products will be connected to the "Tongyi Qianwen" model in the future and undergo a comprehensive transformation. The products involved are not only DingTalk, AutoNavi Maps, Xianyu, Leome and Hema also include core products such as Taobao and Tmall.

"Facing the AI ​​era, all products are worth re-upgrading with large models." Zhang Yong, who "completely transformed" Ali with AI, will lead the development of Alibaba Cloud in the future, which deserves the continued attention of AI industry.

PS AI becomes popular on the Chinese network

Meme expansion is hilarious

In the past week, a group of AI meme extensions became popular on the domestic Internet. It's just stepping on red high-heeled shoes and being held in the arms of Master Xing "Oboi":

Jacky Cheung, who is still talking to the pot, "Eating shit, you", AI's imagination of "outside the camera" is always surprising:

In fact, such a meme is not difficult to make. The core cheat is "Install the overseas version of Photoshop (Beta), and use the Generative Fill tool to expand the image with one click." We have also written a basic tutorial before , interested friends can learn it together, you You will find that AI image expansion is so simple.


no dataCoffee time! Feel free to comment