App analytics firm Appfigures has analyzed download and revenue spikes for the three big consumer AI apps in the wake of recent image-model launches. The headline finding: image-model launches now generate 6.5× more download spikes than traditional chatbot upgrades. The numbers from 28-day windows after each launch are striking. Gemini's Nano Banana (Gemini 2.5 Flash image model, Aug 2025) pulled 22M+ downloads against approximately $181K in estimated gross consumer spending. ChatGPT's GPT-4o image model (Mar 2025): 12M+ downloads, ~$70M revenue. Meta AI's Vibes (Sept 2025, AI video feed): 2.6M downloads, no meaningful revenue.
The number that should make builders sit up: ChatGPT converts at roughly $5.83 per download in the post-launch window. Gemini converts at roughly $0.008 per download. That's a ~700× difference in monetization efficiency. Caveat the data carefully — Appfigures measures *consumer app-store revenue*, which excludes Google One subscriptions, ads, and API spending, all of which feed Google's actual Gemini revenue picture. Subscription-driven apps that route revenue through their own checkout (ChatGPT) show up cleanly; ad-and-cloud-driven apps don't. So the gap is partly real (ChatGPT's subscription muscle is the strongest in the consumer AI space) and partly methodology (the Appfigures lens undercounts Gemini's revenue sources). The 6.5× multiplier on download spikes is the more methodologically robust finding.
For builders shipping consumer AI apps, the takeaway is the structural shift in acquisition. Image-model releases are now the dominant download-driver for the major chatbots, beating model-quality upgrades for the underlying conversation model. That's a real change in user attention: people show up for the new image capability, not for "GPT-4o → GPT-5" or "Gemini 2.0 → 2.5" reasoning improvements. The reasoning-model upgrade cycle isn't the marketing vehicle anymore; the image-and-video model release cycle is. Anyone building a consumer AI app should plan their acquisition story around generative-image moments, not around incremental reasoning gains.
For monetization, the structural lesson is that image generation is a great hook and a bad SKU. ChatGPT works because the image capability is one feature inside a $20/month subscription that users were already buying for the model bundle as a whole. Gemini's image-spike-with-no-revenue suggests that *in the consumer surface*, free image generation isn't a thing people pay for incrementally — they expect it as part of whatever they were already on. If you ship image generation as your monetization hook, you'll get downloads and not revenue. Wrap it inside a feature stack people pay for the bundle of, and conversion follows.