OpenAI launches GPT Image 1.5 after Nano Banana goes viral

Google went bananas — Nano Banana, that is. Now, OpenAI has peeled into the chat.
On Tuesday, the company rolled out a revamped ChatGPT Images experience powered by GPT Image 1.5; the new interface will reach most users right away — though Business and Enterprise users will get it later.

The company says the model is up to four times faster and is also available via the API as GPT-Image-1.5. And the redesigned ChatGPT Images section adds a dedicated Images tab in the sidebar and leans into preset filters and prompt suggestions, perhaps a less obvious way of saying OpenAI wants viral formats to live on its turf. 

In a preview video, OpenAI showed side-by-side comparisons to highlight fidelity gains, including a 1970s Chelsea, London, scene and a mechanic working on a car. And that might be a tell about what OpenAI thinks the real contest is: iterative edits that keep the face, keep the lighting, and keep the world consistent —and don’t quietly swap reality out from under you.

The timing is obvious: Nano Banana Pro’s recent run of praise had sparked chatter about whether Google was starting to — gasp! — outpace OpenAI and change the AI race. Google’s Nano Banana and Nano Banana Pro — the meme-friendly nicknames for Gemini image models that can generate and edit images conversationally — have been soaking up attention for exactly the kind of “make this specific thing, then tweak it without breaking it” magic that turns a tool into a trend.

Google’s own developer docs map the branding to the plumbing: Gemini 2.5 Flash Image for Nano Banana, and Gemini 3 Pro Preview for Nano Banana Pro. And the company’s Nano Banana Pro pitch leans on camera and lighting controls, mockups built from multiple images, and text-heavy outputs such as posters and diagrams. Google has billed Nano Banana Pro as capable of blending complex compositions using up to 14 images while keeping the consistency and resemblance of up to five people, plus improved text for posters, invitations, and context-rich diagrams.

OpenAI is trying to win the same kind of repeatability inside ChatGPT. Its own post frames GPT Image 1.5 as better at preserving branded logos and key visuals across edits and useful for e-commerce teams building full product image catalogs from a single source image. It also says image inputs and outputs are 20% cheaper in GPT Image 1.5 than GPT Image 1. OpenAI’s pricing docs list gpt-image-1.5 at $8 per 1 million input image tokens, $2 per 1 million cached input, and $32 per 1 million output image tokens.

Google, meanwhile, is trying to keep the banana from becoming a trust problem. Its AI Studio model pages say images created or edited with Gemini 2.5 Flash Image include an invisible SynthID watermark, and TechRadar reports Google also uses a visible “Gemini sparkle” watermark for free and Pro tier users while removing the visible watermark for Ultra subscribers and in AI Studio. Google’s Nano Banana Pro pitch, meanwhile, foregrounds camera and lighting controls, mockups built from multiple images, and text-heavy outputs such as posters and diagrams — the practical stuff that keeps an image generator from feeling like a one-trick carnival booth.

Google threw a banana peel onto the AI race track. But OpenAI didn’t faceplant — it sped up.