Image model releases drive 6.5x more app downloads than traditional AI model updates

[kevin1]

A new report from app intelligence provider Appfigures has found that image model releases generate 6.5 times more downloads for AI mobile apps than traditional language model updates, marking a fundamental shift in what drives consumer adoption of AI products. The data reveals that visual capabilities have become the primary growth lever for the major AI platforms, overtaking the conversational model releases and feature updates like voice chat interfaces that drove earlier waves of demand. Google's Gemini saw its download rate increase more than 4x in the 28 days following the release of its Nano Banana image model last August, adding more than 22 million incremental installs over that period. ChatGPT added more than 12 million incremental installs in the 28 days after the introduction of its GPT-4o image model in March 2025, representing approximately 4.5 times more downloads than it generated from the releases of GPT-4o, GPT-4.5, and GPT-5 as language models.The pattern reflects a straightforward consumer psychology dynamic: image generation produces immediate, shareable, visually compelling outputs that are easier for non-technical users to appreciate and demonstrate to others than improvements in reasoning or conversational quality.

When someone generates a striking AI image, they share it. When a model gets better at logical reasoning, the improvement is harder to show in a social context. Meta AI's introduction of its AI video feed Vibes added an estimated 2.6 million incremental downloads in the 28 days after its September 2025 release, following the same visual content pattern on a smaller scale. The download numbers paint a clear picture for AI product teams: visual capability announcements are the most reliable consumer acquisition events available, and the gap between image model releases and language model releases in terms of download impact is large enough to treat as a strategic planning assumption rather than a coincidental data point.

[etfs]

"Improvements in logical reasoning are harder to show socially" is the most polite way to say nobody cares

[johnblockbuster]

My mom still doesn't understand what GPT-5 does but she showed me her AI photo at dinner last Sunday