From AI Portraits to AI Image Expansion: Are Image Applications More Likely to Go Viral?
-
As early as July, Meitu's official Weibo announced the launch of the 'AI Image Expansion' feature, now available on both Meitu App and Wink App for immediate experience. Recently, as Miaoya gradually faded from the internet, AI Image Expansion has taken over the discussion buzz.
Recently, Kuaishou's self-developed text-to-image model 'Kolors' also began internal testing.
In fact, since ChatGPT went viral across the internet, domestic AI-related applications have also emerged in the public eye. However, from Miaoya to Meitu, most consumer-facing applications that have gone viral are image-generation tools.
Did Meitu 'Cut Off' Miaoya, with Kuaishou Following?
'AI Image Expansion is amazing—it directly saves my ruined travel photos.'
'I might actually call the police; the results from AI Image Expansion are both hilarious and absurd.'
'AI Image Expansion did a great job—next time, I won’t use it.'
...
Recently, AI-expanded images from Meitu have become ubiquitous across major social media platforms.
The feature, officially described as using AI algorithms to intelligently predict and generate missing parts based on image context and textures, can expand original photos to wider aspect ratios and broader perspectives.
Currently accessible on Meitu's app homepage, users get five free daily attempts to expand images at 150%, 200%, 250%, or 300% ratios (default 125%).
On Xiaohongshu, the #AI扩图# topic has garnered 21.28 million views, while Douyin shows over 2 million plays. A popular Xiaohongshu post titled "Fainted from AI expansion - your aesthetic sense is too good" received 33k likes.
Public reactions are polarized: while many praise its composition-enhancing capabilities, bizarre outputs like strange clothing additions, extra limbs, unfamiliar faces, or floating heads have also contributed to its virality.
AI image expansion isn't novel - Adobe Photoshop's "Generative Fill" launched in May, and Midjourney's July update introduced more flexible panning expansion. However, Meitu has popularized the technology for Chinese users, becoming the current focus of discussions.
This follows previous AI photo trends like portrait-generating apps "Miao Ya Camera" and "45AI," whose popularity has now been eclipsed by Meitu's controversial expansion feature.
If Meitu's AI image expansion feature has successfully 'intercepted' Miao Ya's momentum, we must note that Kuaishou is waiting in the wings.
In September, Kuaishou began internal testing of 'Kuaishou AI Playful Comments' in short video comments. According to official introductions, this feature leverages Kuaishou's self-developed text-to-image model 'Kolors,' allowing users to generate various styled images through creative text inputs for more engaging interactions in comment sections.
However, discussions about 'Kolors' remain scarce across major social media platforms.
Why are image-generation applications more likely to go viral?
Looking back in recent years, most genuinely viral AIGC products in China seem to emerge in the image-generation field.
'I've tried AI chat tools too, but they often feel dumb and less interesting than expected. Apps like Miao Ya and AI image expansion, though - I've used them all. You frequently get surprising results, whether bizarre or impressive, and they generate good engagement when shared on social platforms,' says 20-year-old Taozi, a social media enthusiast curious about all things new. Compared to conversational AI like ChatGPT, she prefers experimenting with image-generation apps.
'Image-related applications are indeed more readily accepted by users because, at the application level, people's expectations of AI image functions are more utilitarian - users have clearer expectations,' explains Qi Sui, an algorithm engineer at a domestic tech company. He notes technical reasons behind image apps' viral potential: 'Globally, the AI wave of the past decade sprouted from the 2012 ImageNet competition, so AI model structures are better suited for image applications, with superior effects and efficiency compared to text. It wasn't until the 2018 invention of Transformer architecture that text applications became competitive, and even then, they remained relatively distant from consumer-facing applications until the recent breakthrough in generative AI brought them to public awareness.'
In China, text content generation is more challenging than image content generation. "Text training requires large amounts of high-quality data, and Chinese data presents greater challenges in both quality and quantity compared to English data. Especially for conversational AI, which is open-ended and involves more diverse tasks, the demand for data quality and quantity increases significantly," said Qi Sui.
However, in Qi Sui's view, the future direction of AI development lies in multimodal large models combining text and images. "36Kr" has previously mentioned that to get closer to artificial general intelligence, improving AI's understanding of multimodal content such as images, speech, and videos is essential. Compared to single-modal language models focused on text interaction, multimodal models can expand the AI field to broader areas such as recognition and detection.
So, what will be the next AI application to capture widespread attention?