Image AI models are now driving app development, surpassing chatbot improvements

Image model updates are driving AI mobile app growth, generating 6.5 times more downloads than customary model updates, according to a novel report from app intelligence provider Appfigures.

This marks a change from earlier days when the release of novel conversation-enabled models resulted in greater demand, along with novel features such as a voice chat interface.

For example, ChatGPT and Gemini added tens of millions of novel downloads after releasing their image models, Applications found.

In the case of Google Gemini, the launch of the Nano Banana image model resulted in over 22 million downloads within 28 days of the launch of the Gemini 2.5 Flash image model last August. The data showed that after the app was introduced, the number of app downloads more than quadrupled during this period.

Meanwhile, ChatGPT added over 12 million incremental installs within 28 days of rolling out the GPT-4o image model in March last year. That’s about 4.5 times more downloads than the GPT-4o, GPT-4.5 and GPT-5 model releases, Appfigures noted.

Other model releases followed similar trends, although on a smaller scale. Meta AI’s introduction of the AI ​​Vibes video channel resulted in an estimated 2.6 million additional downloads within 28 days of its release in September 2025. (Yes, technically it’s a video model, but ultimately it’s about visual content, not just text.)

However, the report cautioned that additional downloads do not always translate into increased revenue from mobile devices.

Instead, novel versions of image models give people a reason to install the app and try out its improved image generation capabilities. However, this does not mean that they will necessarily convert to paid subscribers. For example, Appfigures noted that Nano Banana generated only $181,000 in estimated gross consumer spending within 28 days of launch, even though it drove a larger enhance in downloads than ChatGPT’s version of the 4o image model.

Meta AI’s launch of Vibes also resulted in additional downloads, but did not result in significant revenue.

Of these three, only ChatGPT converted increased attention into actual dollars.

Appfigures says OpenAI’s 4o image generation model led to estimated gross consumer spending of $70 million in the 28 days after its launch compared to its earlier baseline model.

The company also included DeepSeek in its analysis, but it did not fit the pattern.

While DeepSeek R1 was downloaded 28 million times upon its January 2025 launch, it was not a typical model comparison event. This was a watershed moment for DeepSeek, when it went from relatively unknown to an overnight sensation as the tech industry learned about the techniques it used to train its AI models at a fraction of the cost of its competitors. This case shows how curiosity can drive downloads – although in this case the interest was not linked to the image model.