Nano Banana 2.1 vs GPT-Image-2.5: The New Image Models
Google's Nano Banana 2.1 and OpenAI's Images 2.5 both landed this autumn. What each one does best, where to use it and the prompts that get clean edits.
By the WorkCrafter team · how we write these guides
Two image models changed this autumn: OpenAI released ChatGPT Images 2.5 on September 8, 2026, and Google published Nano Banana 2.1 on October 6. Pick Nano Banana 2.1 for infographics and multi-character edits; pick Images 2.5 for photo-based edits that must leave everything else untouched.
Nano Banana 2.1 at a glance
- Released October 6, 2026, built on Gemini 3.6 Flash, takes text and images as input.
- Where to use it: the Gemini app, Google AI Studio, the Gemini API, Search AI Mode, Flow, Stitch and Google Ads.
- Strong at infographics, text inside images, keeping several characters consistent and changing styles.
- Known weak spots listed by Google: small or long text can blur, and left and right are sometimes confused.
ChatGPT Images 2.5 at a glance
- Released September 8, 2026, for every ChatGPT plan, ChatGPT Work and Codex.
- In the API it comes as GPT-Image-2.5 Flare (the default) and GPT-Image-2.5 Sunburst (slower, for tighter control).
- Strong at editing only what you asked, keeping faces from reference photos recognisable, and transparent backgrounds.
- Images carry C2PA metadata and an invisible watermark.
What's new in each
In Google's own human-preference tests, Nano Banana 2.1 beat Nano Banana 2 and Nano Banana Pro on overall preference, infographic design and every editing category listed, with the widest gap on keeping several characters consistent.
OpenAI says Images 2.5 cuts generation time by up to 50% compared with Images 2.0. ChatGPT also gains Sketch (type @Sketch to draw a rough layout that becomes an image), templates for posters and product shots, comments placed directly on an image, and prompts you can share.
How to use the new image models
- Nano Banana 2.1: open the Gemini app or Google AI Studio, upload your reference images and describe the change in one sentence per element.
- Images 2.5 in ChatGPT: upload a photo, then ask for one change at a time. If you already have a layout in mind, start with @Sketch.
- In the API: use GPT-Image-2.5 Flare by default and switch to Sunburst for final campaign images that need tighter control.
- On WorkCrafter: the image tool runs one prompt through several engines so you keep the best result, then the Media Studio upscales it or removes the background.
Our tips for the new image models
- Name what must stay the same ("keep the face, the logo and the lighting") before you name what changes.
- For infographics, give the exact text in quotes and keep each label short; both models still struggle with paragraphs.
- Make one edit per message. OpenAI says Images 2.5 holds earlier changes more reliably across several small edits.
- Keep the watermark and metadata on images you publish commercially.
More on the new image models and WorkCrafter
- The AI Image Tool: Choosing the Right Engine
- How to Generate AI Images That Actually Look Real
- Can You Sell AI-Generated Content? Rights, Explained
- Open the image generator
Sources on the new image models
- Nano Banana 2.1 — Model Card — Google DeepMind, October 6, 2026. Base model, inputs, distribution channels, preference results, known limitations.
- Introducing ChatGPT Images 2.5 — OpenAI, September 8, 2026. Speed figure, Sketch, templates, availability, Flare and Sunburst, C2PA and watermark.
