SOTA Generative Media Panel — Dumitru Erhan, Shane Gu & Nicole Brichtova, Google DeepMind
Google DeepMind launched NanoBanana 2 Light, a faster cheaper image model approaching frontier quality
“getting kind of that like 3 second latency just unlocks a whole bunch of”
Google DeepMind researchers unveiled NanoBanana 2 Light at AI Engineer, describing it as their fastest and cheapest image model that surpasses the original NanoBanana for generation and editing tasks. The model achieves roughly 3-second latency while approaching the quality of larger frontier models, making it a practical replacement for most use cases. The panel also touched on Omni Thinking and Gemini RL work, signaling broad generative media investment at DeepMind.