Image editing in Gemini just got a major upgrade

Source: DeepMind Date: 2025-10-23 URL: https://deepmind.google/blog/image-editing-in-gemini-just-got-a-major-upgrade/

Summary

Google integrated “Nano Banana” — an image editing model claiming top rank on lmarena.ai’s leaderboard — into the Gemini app. The model’s primary differentiation is identity-consistent photo editing: preserving a person’s or pet’s likeness across edits (clothing changes, background swaps, style transfers) while supporting multi-turn iterative editing. All outputs carry visible watermarks plus SynthID.

Implications

Identity consistency as the consumer editing wedge. The specific focus on making edited photos “look consistently like themselves” targets the failure mode that makes AI photo editing feel uncanny — subtle identity drift. If Nano Banana holds likeness better than alternatives, that’s the feature that converts casual Gemini users into regular photo editing users.

lmarena.ai leaderboard as marketing, not science. “Top-rated image editing model in the world” is a preference ranking on an online arena, not a held-out benchmark. That’s a weaker claim than it sounds — preference evals are gameable and style-dependent. The claim deserves verification against specific editing tasks.

Gemini app as the consumer editing surface. Embedding this into the Gemini app (not just API) means the distribution play is consumer-first. Watch whether this drives Gemini app MAU growth in the same way that ChatGPT’s image generation drove OpenAI consumer metrics.

Watch:

• Third-party evaluations of identity consistency vs. Adobe Firefly and Midjourney’s portrait editing
• Whether multi-turn editing in the Gemini app becomes a stickiness driver
• SynthID’s role as a provenance system — all Nano Banana outputs are watermarked, which matters for platform content policies