The AI app game has fundamentally shifted. According to app intelligence firm Appfigures, image model releases now generate 6.5 times more downloads than traditional model updates. That’s a staggering swing from just a couple of years ago, when new conversational models and features like voice chat were the main draw.
It’s a telling sign of where consumer interest really lies. But here’s where it gets interesting: downloads and revenue aren’t the same thing.
The Download Explosion
The numbers are undeniable. When Google rolled out Gemini’s Nano Banana image model last August, the app saw 22+ million additional downloads in the 28 days that followed. That’s a 4x spike in activity. ChatGPT’s GPT-4o image model in March generated over 12 million incremental installs in the same timeframe, roughly 4.5 times what it saw from its GPT-4o, GPT-4.5, and GPT-5 model releases combined.
Even Meta AI’s Vibes video feed, which launched in September 2025, added an estimated 2.6 million downloads in its first month. The pattern is clear: if you want people to download your Technology app, give them a shiny new way to generate images.
These releases clearly tap into something people want to try immediately. They’re tangible, visual, immediately gratifying. A new language model might sound impressive in a press release, but a new image generator? That’s something you can actually play with right now.
The Revenue Problem Nobody’s Talking About
Here’s the twist that most coverage glosses over: downloads don’t automatically equal money.
Nano Banana drove those massive numbers but generated only an estimated $181,000 in gross consumer spending over its 28-day window. Meta’s Vibes had similar issues, producing additional downloads with effectively zero meaningful revenue attached.
Only ChatGPT actually converted the spike into dollars. OpenAI’s 4o image model led to approximately $70 million in gross consumer spending over that same 28-day period compared to its baseline. That’s the outlier. That’s the one that actually worked.
This gap reveals something uncomfortable for the industry: a download surge doesn’t necessarily mean you’ve built something that makes people want to pay. It means you’ve given them a reason to install the app and kick the tires. Whether they stick around and open their wallets is an entirely different question.
Why DeepSeek Breaks the Pattern
Appfigures looked at DeepSeek R1 as well, but it doesn’t fit neatly into the image model thesis. When DeepSeek released R1 in January 2025, it pulled 28 million downloads. But this wasn’t a case study in image generation dominance. It was something rarer: a genuine cultural moment.
DeepSeek went from relative obscurity to household name overnight, not because of visual content, but because the tech industry realized it had achieved powerful AI capabilities at a fraction of the cost everyone assumed was necessary. Curiosity, not features, drove those downloads.
It’s a reminder that sometimes the story isn’t about what the app does, but about what it means. DeepSeek’s moment was Business news wrapped in a technical achievement. That kind of narrative momentum is harder to manufacture than releasing new features.
What This Actually Means
The shift toward image model releases makes sense from a product perspective. They’re concrete, shareable, and fun in a way that model architecture improvements never will be. They give users an immediate reason to reinstall an app or try a competitor’s version.
But the revenue data suggests the industry is optimizing for engagement theater while missing the harder problem: converting casual users into paying customers. You can drive 22 million downloads and still generate pocket change if those users aren’t interested in subscribing or making in-app purchases.
The winners in this space will be the companies that figure out how to turn curiosity into commitment, not just the ones that can engineer the next viral feature.
