Edited by | Yan Zheng

Recently, Huawei terminal BG Bruce Lee posted a blog to show the AI photo enhancement function of Mate70 Pro+, and was eager to ask you whether you are satisfied with the AI cloud enhancement effect.

picture

In order to facilitate everyone to see the effect, you can click on the large image below to see the AI enhancement effect.

On the left is the original image, and on the right is the AI enhancement image

On the left is the original image, and on the right is the AI enhancement image

However, a curious netizen asked a question that Bruce Lee was very concerned about in the comment area: Xiao Long, high rate can not see clearly the word AI (enhanced), it is best to be natural, can not be made into a fake.

picture

Bruce Lee reply: Text with AI enhancement is too easy to flip over, and it is still a problem in the industry.

picture

1. What is AI Cloud Enhancement

First, let's review the science popularization here, what is Huawei AI cloud enhancement?

This must first mention the AI cloud camera function, which was launched by Huawei in October 2023.

At that time, Yu Chengdong gave an image example --

"In the past, when we took a group photo, we would find it difficult to make the details of all people's faces clearly visible after the image was enlarged." Huawei Mate60 Pro, Huawei MateX5 and other mobile phones have launched stunning AI cloud enhancement functions on photos, capturing and restoring tiny details hidden in images through the cloud's 100 million level parameter large model algorithm, bringing a clearer and more delicate portrait experience.

Whether it is a large group photo of 50 people or a single photo, the skin of the portrait is more transparent and delicate, and everyone is the protagonist. Even if the photo is enlarged and the composition is cropped twice, the details are still very clear. Welcome to upgrade your experience!"

picture

Just in 2024, at the end of November, Huawei released the Mate70 series, which continued to upgrade the image technology: Mate 70 series image photography is equipped with the "red maple primary color camera sensor" for the first time in the industry, so that the picture presents a richer sense of layer and reality, but also brings the epic enhancement of AI cloud camera function - long focus cloud large model.

Naturally, the strongest Mate flagship Mate70 series in history has also made innovations in AI cloud enhancement functions, for any shooting scenes with a telephoto of 10x and above (not limited to photos of people), Al cloud enhancement can intelligently improve photo clarity and immediately display high-definition texture.

Huawei's official website describes this function. The AI cloud enhancement function can intelligently analyze photos taken by users, capture and restore tiny details hidden in images based on the cloud's hundred-million-level parameter large model algorithm, and improve photo clarity. This feature supports:

• Portrait photos, flower photos, pet photos taken in photo mode and portrait mode (support 20X or less).
• Photos taken in Macro mode: In camera mode, close-up objects will automatically enter macro, or select Super Macro mode in the Camera [More] page to shoot objects within 20cm.
• Supports photos of any scene at 10X or above in photo mode (except photos taken in moon mode).

The current mobile devices supported by AI cloud enhancements are as follows:

• Portrait photos: Mate70 Series, Mate X6, Mate XT, Pura 70 Series, Mate 60 Series, Mate X5, Pocket 2 and nova 12 Ultra
• Floral and macro photos: Mate XT, Pura 70 Series and Mate 70 Series
• Pet photos, any scene photos of 10X and above: Mate 70 series

Why is it so hard for AI to generate text on photos?

As Bruce Lee said, at present, do not say that high-magnification fuzzy text AI repair, even if let AI add text on ready-made pictures, it is difficult to do it correctly.

Why is that?

First, the AI tools currently on the market for image generation or photo restoration do not specifically learn the complexity and accuracy of language during training. Therefore, when the AI image tool is faced with the task of generating text, it naturally does not complete the task well.

Second, there is a serious shortage of training data for AI text generation. In the process of AI training, there is not enough "label" of text samples, because there is no precise "label" of corresponding text in its basic corpus, or the trainer intentionally avoids images containing text during training to prevent the generation of inappropriate content or copyright infringement, which will lead to insufficient basic data.

In addition, there are difficulties in semantic understanding. The generation of text involves a deep understanding of natural language, which is also a relatively complex problem.

picture

Of course, what Teacher Li mentioned is to let AI fix the problem of fuzzy text in high-rate photos, Teacher Li replied: easy to flip over, that means Huawei should have been studying this aspect of the content, looking forward to Huawei's new results!

Reference link:

https://zhuanlan.zhihu.com/p/10160709558

https://www.zhihu.com/question/628375704/answer/3343109727

To learn more about AIGC, please visit:

51CTO AI.x community

https://www.51cto.com/aigc/