Google supercharges image editing in Gemini

The biggest headache in AI photo editing has always been the face-swap roulette. Ask an AI to change your shirt, and it happily hands you an entirely new face and a completely different room.

Google is fixing that. The company just rolled out Gemini 2.5 Flash Image across the Gemini app, Gemini API, Google AI Studio, and Vertex AI.

If that capability rings a bell, it is because this was the mystery model tested under the pseudonym nano-banana on the LM Arena leaderboard a few weeks ago, where it crushed the competition and took the number-one spot over Flux-Kontext by a massive margin.

Nano-banana drops the disguise

The new model, developed by Google DeepMind, tackles the Achilles' heel of AI image tools: character consistency.

Nicole Brichtova, product lead for visual generation at DeepMind, noted that the model is fine-tuned specifically to preserve likeness across edits. You can drop in a selfie and ask for a 1960s haircut, or dress your chihuahua in a tutu, and the subject still unmistakably looks like them.

Key features rolled out include:

Multi-turn edits: You can upload a photo of an empty living room, ask Gemini to paint the walls blue, and then keep the conversation going to add a bookshelf and a coffee table without destroying the room's geometry.

Photo blending: Upload a photo of yourself and a separate photo of your dog, and prompt Gemini to place both of you together on a basketball court.

Style transfer: You can pull patterns from butterfly wings and apply them directly to a dress design, leveraging the model's multimodal reasoning and real-world knowledge.

Google is also undercutting rivals on price. The model runs at $0.039 per image via API, slightly cheaper than OpenAI's gpt-image and BFL's Flux-Kontext. Edited photos can also be fed straight back into Gemini to generate videos. Every generation includes both a visible badge and invisible SynthID watermarking in the metadata.

Why it matters

We think this is one of Google's sharpest product moves all year. Instead of chasing hyper-stylized novelty art, DeepMind focused on solving the frustrating workflow flaws that make everyday photo editing annoying. If you are a creator, designer, or casual user staging home renos, it is definitely worth testing right away.

Google also desperately needs this hit to close the engagement gap with OpenAI. ChatGPT boasts over 700 million weekly active users, while Google CEO Sundar Pichai reported 450 million monthly active users for Gemini in July. OpenAI saw massive user surges when GPT-4o sparked viral image generation trends, and Google wants that same viral fuel for Gemini.

While Google's strict safeguards prevent non-consensual imagery—steering clear of the controversies currently hitting platforms like xAI's Grok—the lingering challenge is moderation. Invisible watermarks like SynthID rarely stop deceptive images from fooling people casually scrolling on social media feeds.