I tested Nano Banana 2.1 against ChatGPT, and it was close

Google has given its Nano Banana AI image generator another upgrade, and the latest version, Nano Banana 2.1, is arriving inside Google Gemini in an increasingly crowded field of tools that can turn a few sentences into remarkably convincing pictures.

<![CDATA[ <article> <p>Google has given its <a href="https://www.techradar.com/ai-platforms-assistants/gemini/5-prompts-that-show-how-powerful-nano-banana-2-is">Nano Banana</a> AI image generator another upgrade, and the latest version, Nano Banana 2.1, is arriving inside <a href="https://www.techradar.com/computing/artificial-intelligence/what-is-google-gemini">Google Gemini</a> in an increasingly crowded field of tools that can turn a few sentences into remarkably convincing pictures. The promise is familiar, with sharper details, better instruction-following, more accurate text, and greater consistency across images. The real challenge is finding something that makes it meaningfully better than the competition.</p><p>Nano Banana's main rival, <a href="https://www.techradar.com/news/chatgpt-explained">ChatGPT</a>'s image generator which is called<a href="https://www.techradar.com/ai-platforms-assistants/chatgpt/chatgpt-images-2-5-is-out-ive-been-testing-it-for-24-hours-and-these-are-the-3-new-features-youll-actually-use"> Images 2.5</a>, is hardly sitting around waiting to be overtaken. OpenAI has made considerable progress with realistic lighting, complicated compositions and text rendering, turning image generation into one of ChatGPT's most entertaining and genuinely useful features. With both companies pushing the technology forward, I wanted to see whether Google's latest upgrade had managed to pull ahead.</p><p>If you've got a Gemini account you'll find that it should have been upgraded to use Nano Banana 2.1 by now. To test the two image generators against each other I gave Gemini and ChatGPT identical prompts designed to test different abilities. I compared the resulting images side by side, with ChatGPT on the left and Gemini on the right. The results were surprisingly close, although the two generators often approached the same instructions with different artistic instincts.</p><h2 id="storms-within-storms">Storms within storms</h2><a href="https://cdn.mos.cms.futurecdn.net/dTv5JTW7Zxtf8dULExWa7j.png"><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:2021px;"><p class="vanilla-image-block" style="padding-top:37.85%;"><img id="McGwqMNq8eZQNcTC5RKux" name="Nano Banana 2.1 1" alt="Nano Banana 2.1 / ChatGPT" src="https://cdn.mos.cms.futurecdn.net/McGwqMNq8eZQNcTC5RKux-1920-80.png" mos="" align="middle" fullscreen="" width="2021" height="765" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">A tiny Victorian greenhouse inside a glass snow globe during a thunderstorm. ChatGPT on the left, Gemini on the right. </span><span class="credit" itemprop="copyrightHolder">(Image credit: ChatGPT/Gemini)</span></figcaption></figure></a><p>My first test was designed to challenge the models' understanding of light, glass and reflections, three things that can quickly expose the weaknesses of an AI-generated image. I wanted a scene with multiple transparent surfaces, complicated lighting and enough tiny details to make the results worth examining closely. The prompt was, <em>"Create a photorealistic image of a tiny Victorian greenhouse inside a glass snow globe during a thunderstorm, with realistic reflections, refractions, and lighting."</em></p><p>Both generators produced beautiful images that would look perfectly at home on the cover of a fantasy novel. ChatGPT's version featured an elaborate, golden-lit Victorian greenhouse inside a large glass sphere, sitting on a polished wooden base beside a rain-covered window. Lightning illuminated the storm outside, while the globe contained a surprisingly detailed miniature world, complete with plants, pathways and delicate architectural flourishes.</p><p>Nano Banana 2.1 went for something slightly more restrained. Its greenhouse looked more like an actual miniature structure, with dark metal framing, lush greenery and warm interior lighting that contrasted beautifully with the blue-gray storm outside. The globe's reflections were particularly convincing, and the lighting felt a little more natural, although the wet tabletop was an odd choice given that the scene appeared to be indoors.</p><h2 id="titan-calling">Titan calling</h2><a href="https://cdn.mos.cms.futurecdn.net/gnKbSTrFS82YmDTfbuh93k.png"><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:1538px;"><p class="vanilla-image-block" style="padding-top:66.58%;"><img id="ETDuUfP4uD2L9Tj6P6mD73" name="Nano Banana 2.1 2" alt="Nano Banana 2.1 / ChatGPT" src="https://cdn.mos.cms.futurecdn.net/ETDuUfP4uD2L9Tj6P6mD73-1920-80.png" mos="" align="middle" fullscreen="" width="1538" height="1024" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">ChatGPT on the left, Gemini on the right. </span><span class="credit" itemprop="copyrightHolder">(Image credit: ChatGPT/Gemini)</span></figcaption></figure></a><p>Text has historically been one of the most embarrassing weaknesses of AI image generators. Letters would melt into one another, words would become nonsense and asking for an exact phrase could produce something that looked like an alien language. For the second test, I wanted to see whether either model could handle a fairly demanding design brief without mangling the typography.</p><p>The prompt was, <em>"Design a retro-futuristic travel poster for Saturn's moon Titan with the exact text 'TITAN IS CALLING,' 'Seven nights beneath the rings,' and 'Starting at $2,499.'"</em></p><p>ChatGPT created a luxurious travel advertisement that looked like something a very wealthy space tourist might encounter in an interplanetary airport. Saturn dominated the sky above a beautiful orange landscape, while futuristic resorts and a couple enjoying drinks suggested that Titan had become the solar system's most exclusive vacation destination. One thing I didn't like was the gap it left between the $ and the 2 in its text rendering.</p><p>Google took a more cartoonish approach with a colorful vintage science-fiction poster featuring rockets, astronauts, icy mountains, and a futuristic base. It was actually more realistic about how the moon is not at all hospitable but still had the optimistic quality of a 1950s space-travel advertisement, complete with bold lettering and a slightly weathered finish. Crucially, it also reproduced the requested text correctly, including the price.</p><h2 id="fox-library">Fox library</h2><a href="https://cdn.mos.cms.futurecdn.net/BHuMoDeLirr7qXBMDNvyNk.png"><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:1570px;"><p class="vanilla-image-block" style="padding-top:48.73%;"><img id="hC46UYA5ymtsxRm3o5vd73" name="Nano Banana 2.1 3" alt="Nano Banana 2.1 / ChatGPT" src="https://cdn.mos.cms.futurecdn.net/hC46UYA5ymtsxRm3o5vd73-1920-80.png" mos="" align="middle" fullscreen="" width="1570" height="765" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">ChatGPT on the left, Gemini on the right. </span><span class="credit" itemprop="copyrightHolder">(Image credit: ChatGPT/Gemini)</span></figcaption></figure></a><p>Character consistency is another tricky problem for AI image generators, particularly when the same character needs to appear in several scenes. A fox might suddenly change its appearance, lose an accessory, or acquire an extra limb between panels. I decided to test this with a simple illustrated story that required both visual continuity and a coherent sequence of events.</p><p>The prompt was, "<em>Create a four-panel comic showing an orange fox wearing a green scarf discovering a magical key, unlocking a door, and entering a library. Keep the fox identical in every panel."</em></p><p>ChatGPT produced four almost 3D illustrations. The lighting seemed more computer-generated, though the fox remained recognizable throughout as requested. The style leaned toward children's book illustrations rather than a conventional comic.</p><p>Nano Banana 2.1 delivered something much closer to a traditional comic strip, with bold outlines, clearly defined panels, and a consistent cartoon fox. Its version began with the fox finding a key in the forest, then examining it, unlocking a door in a tree, and finally entering a magical library. The storytelling was clearer, and the character's appearance stayed impressively consistent, right down to its green scarf and white-tipped tail.</p><p>Google edged ahead here because its four panels felt like parts of the same illustrated story. ChatGPT's individual images were more elaborate, but the fox's proportions and presentation varied slightly, and the sequence felt more like four separate illustrations.</p><h2 id="dragon-on-a-desk">Dragon on a desk</h2><a href="https://cdn.mos.cms.futurecdn.net/DVay8MdRChCAimZfzrHDkj.png"><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:2055px;"><p class="vanilla-image-block" style="padding-top:37.23%;"><img id="Sof3KJnSPW5A6bGSBtMR93" name="Nano Banana 2.1 4" alt="Nano Banana 2.1 / ChatGPT" src="https://cdn.mos.cms.futurecdn.net/Sof3KJnSPW5A6bGSBtMR93-1920-80.png" mos="" align="middle" fullscreen="" width="2055" height="765" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">A desk with a dragon sketched in a notebook. ChatGPT on the left, Gemini on the right. </span><span class="credit" itemprop="copyrightHolder">(Image credit: ChatGPT/Gemini)</span></figcaption></figure></a><p>For my final test, I wanted to see how accurately the models could follow a complicated collection of specific instructions. AI image generators have a habit of overlooking small details when a prompt contains several independent requirements. Getting everything into the picture is one challenge, while getting the exact numbers and arrangements right is another.</p><p>The prompt was, <em>"Create a realistic overhead photograph of a wooden desk with a notebook showing a dragon, a compass pointing north, a pocket watch showing 3:15, exactly three red wax seals, and a blue gemstone."</em></p><p>ChatGPT produced an attractive overhead photograph of an antique wooden desk, with an open notebook containing an intricate dragon illustration. The compass, pocket watch, three red wax seals, and bright blue gemstone were all present, and the composition looked carefully arranged. </p><p>Nano Banana 2.1 also included the requested objects, but the notebook featured a smaller, cleaner dragon illustration, while the wax seals, gemstone, compass and watch were arranged around it. Google's result looked more like a casually assembled collection of historical objects, but it did make a common AI error and got the time wrong, setting it for 3:00, not 3:15.</p><div style="min-height: 250px;"> <div class="kwizly-quiz kwizly-OdR6be"></div> </div> <script src="https://kwizly.com/embed/OdR6be.js" async></script><h2 id="approaching-parity">Approaching parity</h2><p>You can access Nano Banana 2.1 in Gemini if you're a Google AI Plus, Pro or Ultra subscriber as it is the new default model. Free users will still be getting the older Nano Banana 2 model.</p><p>Nano Banana 2.1 is an excellent image generator, but ChatGPT is already so capable that Google's improvements don't necessarily translate into an obvious advantage. Both tools produced attractive, detailed images from relatively short prompts, and neither suffered the glaring visual mistakes that used to make AI-generated pictures instantly recognizable.</p><p>That said, different prompts, styles, and repeated generations could easily produce different results. What these tests do suggest is that the leading AI image generators have reached a remarkably high level of capability and are approaching parity in many everyday creative tasks.</p><p>That makes the competition more interesting, even if it makes choosing between them harder. The differences increasingly come down to taste and interpretation. Nano Banana 2.1 may be Google's latest impressive upgrade, but ChatGPT has made sure it won't be claiming the image-generation crown without a fight.</p> </article> ]]>