Xiaomi already has its own generative AI service but for months it was limited to the Chinese market only. During MWC Barcelona 2024, Xiaomi promised that their AI service will be offered to global users in stages starting from the first quarter of 2024 and this week, users in Malaysia can finally try it.
There are three main AI features on Xiaomi HyperOS namely AI Eraser Pro, AI Expansion and AI Portrait. These three services are said to be free forever because there are absolutely no plans to charge a subscription price. This feature is seen available in Xiaomi HyperOS 1.0.11.0.UNCMIXM with Mi Gallery app version 3.6.2.10-global.
AI Eraser Pro
Like Photo Assist on Galaxy AI for Samsung devices with One UI 6.1, AI Eraser Pro is a feature that can remove or add objects in existing photos. In the Gallery application, select a picture and press the pencil icon then select Create. Then select Erase and there will be an option to Manually, Remove object, Remove lines, Remove people or Remove Shadow to make it easier for the user to do any kind of editing.
AI Expansion
This is like a generative fill in Photoshop where with the crop feature, the user's image can be filled with the background and if the human picture is half the body Xiaomi AI can also add legs. Similar to AI Eraser Pro, press the pencil icon on the picture you want to edit, just don't press Create but select Crop. Then press AI expansion and select a size such as Auto or other display ratios such as 1:1, 16:9 and more. The AI will take about 5-10 seconds to fill in the empty spaces and add what is appropriate.
AI Portrait
The feature of generating new images by itself by training the AI using its own photos has been offered since last year. Usually users need to upload 5-10 pictures and wait for their turn. This method usually requires users to pay for each generation or subscribe monthly to generate multiple times. On Xiaomi HyperOS, the same concept is used but requires 25 to 35 pictures of the same user with different clothes and different facial styles.
Xiaomi's AI system will then upload the picture to their server and the user will have to wait from 20 minutes to 2 hours, if many people use it it may take longer like me who has to wait 4 hours. After completion, around 8 images are provided, each image is given an option to download the full resolution version. If not satisfied, users can enter prom to generate pictures with different styles and themes which takes around 1 minute to 30 minutes.
Although this feature is free, users are limited to generating only 20 photos per day. There is no option to generate more photos or the option to purchase additional credits. The output is quite interesting and realistic because my face looks similar to the real image.
Xiaomi does not share the language model technology used. However, everyone knows Xiao AI is Xiaomi's main AI which has also been updated with generative AI. On Github Xiaomi has listed the MiLM-6B which supports 6.4 billion parameters but it is intended for Xiao AI so not sure which AI technology this image editing generative AI uses.