New AI image models boost downloads but struggle to convert revenue

A new report from app intelligence provider Appfigures reveals that releasing image-generation models is a highly effective user acquisition strategy for mobile artificial intelligence applications. According to the report, image model releases generate 6.5x more downloads than traditional model updates. This trend marks a shift from earlier phases of mobile AI adoption, where conversational updates and features like voice chat drove user demand. Major technology companies have recently seen massive spikes in incremental installs following visual model releases.

For Google’s Gemini, the release of its image model Nano Banana drove an additional 22+ million downloads in the 28 days following the introduction of the Gemini 2.5 Flash image model last August. This launch lifted the app’s downloads by more than 4x over that period. Meanwhile, OpenAI’s ChatGPT added more than 12 million incremental installs in the 28 days after the introduction of its GPT-4o image model in March of last year. That represents roughly 4.5x more downloads than it saw for its GPT-4o, GPT-4.5, and GPT-5 model releases. Other releases followed similar trends on a smaller scale; 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.

However, the report cautions that additional downloads do not always translate into increased mobile revenue. While new image capabilities give users a reason to install an app and test its features, they do not guarantee conversion into paying subscribers. The financial outcomes of these releases vary widely:

• OpenAI (ChatGPT): The GPT-4o image-generation model led to an estimated $70 million in gross consumer spending over the 28 days after its launch, compared with its prior baseline.
• Google (Gemini): Despite generating a larger download spike than ChatGPT, the Nano Banana image model drove only an estimated $181,000 in gross consumer spending during its 28-day post-release window.
• Meta (Meta AI): The launch of the Vibes video feed led to additional downloads but generated no meaningful revenue.

The report also highlights DeepSeek R1 as an outlier that did not fit this visual-driven pattern. Following its January 2025 release, DeepSeek R1 drove 28 million downloads. However, this spike was driven by industry curiosity regarding the techniques used to train its AI models at a fraction of the cost of competitors, rather than an image model release. This case demonstrates that viral interest can stem from factors beyond visual feature updates.

Why it matters

AI companies are increasingly using visual capabilities as a hook for user growth, but the data suggests that monetization remains a distinct challenge that is not automatically solved by viral install numbers.