Overview
Image-generation model releases are now the primary growth driver for AI mobile apps, generating 6.5 times more downloads than traditional chatbot model updates, according to a new report from app intelligence provider Appfigures. The data shows a clear shift: visual model launches produce larger download spikes than conversational upgrades, but most developers fail to convert that traffic into revenue.
The data
Appfigures tracked download and revenue patterns around major model releases for ChatGPT, Gemini, and Meta AI. Key findings:
- ChatGPT: The GPT-4o image model (March 2025) drove an estimated 12 million incremental installs in the 28 days after launch — roughly 4.5 times more than the combined GPT-4o, GPT-4.5, and GPT-5 model releases. It also generated an estimated $70 million in gross consumer spending over the same period, compared with its prior baseline.
- Gemini: The Nano Banana image model (August 2025, part of Gemini 2.5 Flash) drove 22+ million incremental downloads in the 28 days after launch — more than four times the app's baseline download rate. However, it generated only an estimated $181,000 in gross consumer spending during that window.
- Meta AI: The Vibes AI video feed (September 2025) added an estimated 2.6 million incremental downloads in the 28 days after release, but produced no meaningful revenue.
DeepSeek R1 (January 2025) drove 28 million downloads, but Appfigures notes this was an outlier — DeepSeek's breakout moment driven by industry curiosity about its training techniques, not an image model release.
The revenue gap
The report highlights a persistent problem: visual model launches generate large download spikes, but fewer than 12% of developers monetize that traffic through in-paint upsells or enterprise SaaS tiers. The delta between virality and lifetime value (LTV) is widening. Only ChatGPT turned its image model launch into significant revenue — likely due to its existing subscription base and in-app purchase infrastructure. Gemini's Nano Banana and Meta's Vibes both produced download surges without corresponding revenue.
When to use it
For developers building AI consumer apps, the takeaway is clear: image model launches are the most effective user-acquisition lever available, but they require a monetization strategy beyond the download spike. In-paint upsells, premium image generation tiers, or enterprise SaaS offerings can capture value from the traffic. Without such mechanisms, the downloads remain a vanity metric.
Bottom line
Image models are the new growth engine for AI apps, but the revenue gap between download spikes and actual spending