Nano Banana 2 vs GPT Image 2: I Tested 10 Identical Prompts
I tested Nano Banana 2 and GPT Image 2 on 10 identical prompts — portraits, brand campaigns, infographics. Side-by-side results and a clear winner.
I’m a sucker for image generation models and how they advance. I remember how much time I spent in Canva years ago, manually creating visuals for Instagram for my previous startup. And then the early AI image generation models came out and they were both amazing and frustrating at the same time. Scrambled text. Unnatural-looking images that screamed “AI made this”. Weird placements that broke the whole structure of the post. You know what I’m talking about. I’m sure you have your own list of frustrations.
So if you now look at where we are and compare it to those early days, it's wild how fast this moved.
I've been tracking it for a while, writing about each iteration as it came out, from Nano Banana to Nano Banana Pro to Nano Banana 2 with 58 infographic prompts I built for myself.
And since we launched Amplifiers, image generation has been built in from day one, so you can generate images right inside Claude, first with Nano Banana 2, and more recently with GPT Images 2 too.
Now that both models are available in the same place, I’ve been wanting to put them side by side on the same prompts.
Not the way I usually switch between them, where one model disappoints me and I try the other. A proper experiment. Multiple use cases, same prompts, same conditions, just to see where each one shines and where it falls short.
And today I finally sat down and did it.
The short answer
If you just want the verdict: GPT Images 2 (you’ll also see it called GPT Image 2) won 5 of my 10 tests, and the other 5 were ties. Nano Banana 2 didn’t take a single round outright.
GPT is better at character consistency from a reference photo, realism, and dense editorial layouts like brand boards and landing pages. Nano Banana 2 leans cleaner and more illustrative, which works better when you want a minimal, textbook-style result.
And every test below includes the exact prompt I used, free to copy.
In this article
For each test, you’ll see the exact prompt I used, both outputs side by side, and my verdict:
Test 2: Turning a pet phone photo into a Vogue-style editorial
Test 5: Realistic AI newspaper mockup with your own branding
Let’s begin.
Test 1: Headshot collage with character consistency
Prompt:
Use amplifiers to create a high-end portrait collage using a headshot I will upload as the identity reference. Maintain strong character consistency across all variations. The person must remain clearly recognizable in every frame with the same facial structure, skin tone, eye shape, hairstyle, and overall identity.
Generate a cinematic multi-panel collage featuring the same person wearing different professional and modern headsets across multiple scenes and moods.
Include a mix of: sleek corporate LinkedIn-style portraits, cinematic studio lighting, podcast/interview setup, Netflix character poster, dark moody cyberpunk lighting, Wired magazine portrait photography.Nano Banana 2:
GPT Images 2:
Verdict: The difference here is night and day. The images Nano Banana generated from my reference photo don’t look like me. Not even close. GPT Images 2, on the other hand, kept my face, my hair, my eyes, even the little mole next to my eyebrow. It also preserved my head position from the selfie I uploaded. The clarity, the fidelity, and how natural it all looks is on a completely different level. For character consistency from a reference image, GPT wins this one by a mile.
Test 2: Turning a pet phone photo into a Vogue-style editorial
Prompt:
Use amplifiers to transform the photo I upload of my cat into a high-fashion editorial photoshoot for a luxury pet fashion magazine. Keep the transparent plastic wrapping as avant-garde styling. Use cinematic studio lighting, shallow depth of field, ultra-realistic fur texture, elegant composition, and Vogue-style fashion photography.Nano Banana 2:
GPT Images 2:
Verdict: GPT Images 2 has noticeably higher saturation, which sometimes makes it feel over-processed. The plastic wrapping looks more convincing, the fur texture feels more realistic, and Toby just looks better in the GPT version. Nano Banana’s output looks over-smoothed in comparison. They’re closer on this one than on the headshot test, but GPT still edges it out on realism.
Test 3: Recipe blog post to cookbook-style visual guide
Prompt:
Use Amplifiers to turn this recipe: https://iamafoodblog.com/fluffy-japanese-pancakes-souffle-pancake-recipe/ into a premium cookbook-style step-by-step visual guide for fluffy Japanese pancakes. Include elegant food photography, ingredient callouts, numbered cooking steps, clean editorial layout, soft natural lighting, and realistic pancake textures throughout.Nano Banana 2:
GPT Images 2:
Verdict: Both versions are good on this one. But I still prefer GPT’s. The text placement in the ingredients and step-by-step images feels more natural for a cookbook. In the final pancake image, the maple syrup and butter placement looks more realistic in GPT’s version. In Nano Banana’s, the syrup container sits on the table at the same size as the matcha mug, which looks off.
Test 4: Premium brand campaign board from a product photo
Prompt:
Use amplifiers to create a hyper-realistic, ultra-detailed brand campaign board for Hairburst gummy vitamins (I'll upload a photo) designed like a premium creative agency presentation for a modern wellness brand. The image should feel like a luxury campaign showcase or editorial moodboard, combining multiple visual sections into one cohesive composition.
Use a clean editorial grid layout with a mix of hero product photography, close-up gummy texture shots, lifestyle wellness imagery, packaging mockups, ingredient highlights, infographic-style benefit callouts, typography sections, and social-media-inspired product visuals. The composition should feel curated, balanced, and visually rich, similar to a high-end Behance branding project or launch campaign presentation.
Center the visual identity around glossy pink Hairburst gummies, premium supplement bottles, healthy hair and wellness aesthetics, and ingredient storytelling focused on biotin, selenium, and zinc. Include elegant supplement facts, benefit snippets, and modern wellness branding integrated naturally into the layout.
Use a warm, premium color palette with soft pinks, creamy whites, pastel peach tones, light turquoise accents, and subtle beige neutrals. The lighting should feel soft and natural, with realistic shadows, reflections, and tactile textures throughout.
Add microdetails like sugar-coated gummies, spilled vitamins, condensation on bottles, realistic packaging textures, soft reflections, subtle ingredient illustrations, and polished infographic-style design elements. Avoid sterile pharmaceutical aesthetics, childish candy visuals, dark dramatic lighting, or cluttered branding.
The final image should feel modern, premium, aspirational, and highly shareable, like a luxury wellness campaign created by a top-tier creative agency. Ultra-detailed, 4K.Nano Banana 2:
GPT Images 2:
Verdict: I can’t believe these two visuals came from the same prompt. The GPT version looks like something a creative agency would present to a client. The fonts, the colors, the editorial grid layout, the product photography, it all comes together as a premium, aspirational wellness campaign (as per the prompt instructions). Nano Banana’s version isn’t bad. But when the prompt asks for luxury and premium, GPT delivers that feeling in a way that Nano Banana doesn’t come close to matching here.
Test 5: Realistic AI newspaper mockup with your own branding
Prompt:
Use Amplifiers to create a hyper-realistic mockup of a printed AI newsletter or newspaper called “AI News” with the tagline “Your Daily Piece of AI News.” Design it like a modern editorial tech newspaper focused on practical AI blending the visual style of Monocle, business newspapers, and creator economy media.
The newspaper should feature multiple AI news stories, practical tutorials, sidebars, charts, mini headlines, and editorial sections arranged in a realistic printed newspaper layout with folds, paper texture, ink imperfections, and natural lighting.
The main featured story should prominently showcase:
“Master Claude Like the Top 1% of AI Users” with AI BLEW MY MIND naturally integrated as the featured practical AI newsletter. Include believable editorial copy suggesting that readers use AI blew my mind to learn advanced Claude workflows, save hours at work, improve productivity, and move their business forward through practical AI tutorials and step-by-step guides.Nano Banana 2:
GPT Images 2:
Verdict: This is the closest matchup so far. Both models produced solid newspaper mockups with realistic layouts, paper texture, and believable editorial structure. I’d call this one a tie.
Test 6: SaaS landing page mockup from an article URL
Prompt:
Analyze my article: https://aiblewmymind.substack.com/p/claude-image-generation and then use the Image Generation Amplifier to create a hyper-realistic flat 2D landing page design for it, showcasing how Claude can now generate stunning visuals using Nano Banana and GPT Images with Amplifiers.
The result should look like a real modern SaaS website screenshot, not a 3D mockup or Behance presentation. Use a clean front-facing layout with realistic web design sections, modern typography, polished UI components, beautiful spacing, subtle gradients, and professional SaaS aesthetics.
Present it as a polished vertical landing page showcase with multiple scrolling sections visible together, similar to a premium Behance or Dribbble case study.
Show examples of AI-generated visuals inspired by the article, including infographics, brand systems, social media graphics, product mockups, technical posters, and landing page concepts generated directly inside Claude.
The overall vibe should feel futuristic but approachable, built for creators, marketers, founders, and non-designers who want professional visuals from AI without switching tools.
Core positioning: “Turn Claude into your creative studio.” / “Generate world-class visuals directly inside your chat.” Nano Banana 2:
GPT Images 2:
Verdict: GPT did a better job here. The result feels more realistic and looks more like an actual landing page. Nano Banana’s looks more like a collection of drawn elements arranged together than a real web page. Neither is perfect, but GPT’s version is stronger. With some iteration, GPT’s output could get close to something you’d pass to a coding agent to start implementing.
Test 7: Educational Instagram carousel from a blog post
Prompt:
Analyze this article I wrote: https://aiblewmymind.substack.com/p/how-to-build-voice-ai-agent-elevenagents and use the Image Generation Amplifier to create an educational Instagram carousel.
The carousel should explain how easy it is to create a voice AI agent with ElevenAgents, step by step, based on the article. Make it practical, clear, beginner-friendly for non-technical people, and outcome oriented for business owners and execs.
Include my branding naturally somewhere on each slide, such as a small AI BLEW MY MIND logo or “aiblewmymind.substack.com” in a corner. Make it feel like an educational carousel that teaches people how to build the voice AI agent using Amplifiers.
Format it as a polished Instagram carousel with multiple 4:5 slides, consistent layout, readable typography, and a final CTA slide encouraging people to read AI BLEW MY MIND for practical AI workflows.Nano Banana 2:
GPT Images 2:
Verdict: Neither is perfect. Both models placed the footer in different styles across slides when it should stay the same. Nano Banana even printed a raw color hex code on one slide, and GPT shifted my accent color from purple to blue. The step numbers are also inconsistent in size across slides for both models, which breaks the visual rhythm of a carousel. This is exactly where a dedicated carousel skill would help, since my current brand skill is built for documents, not Instagram slides.
Test 8: Full brand identity system from a single logo
Prompt:
Use the Image Generation Amplifier to create a complete visual identity system for AI BLEW MY MIND, presented as a single large-scale brand board composition based on the logo I’ll upload. The final result should feel like a professional creative direction board that combines branding, editorial design, UI inspiration, social media direction, typography, color systems, and visual storytelling within one cohesive image.
The composition should feature multiple visual sections arranged in a clean editorial grid layout, including logo variations, typography systems, color palettes, Instagram carousel examples, newsletter covers, landing page concepts, graphic motifs, UI cards, content thumbnails, illustration direction, photography treatments, visual patterns, icon systems, social media mockups, AI workflow graphics, editorial layouts, and brand application examples.
Include a dedicated section showing the Image Generation Amplifier in action, featuring realistic before-and-after AI image examples, prompt workflow UI, and generated visuals created through the amplifier. Make it feel like a real product workflow integrated into the AI BLEW MY MIND ecosystem.
The overall aesthetic should feel modern, polished, editorial, and highly curated, similar to a premium Behance branding case study or high-end creative agency presentation.Nano Banana 2:
GPT Images 2:
Verdict: Nano Banana went for a more editorial layout with distinct grid sections, while GPT produced a cleaner, more structured grid with better text legibility and more realistic UI mockup elements. Nano Banana looks more like a mood board, more artistic. GPT looks more like a Behance case study, more production-ready. Neither is what I'd go with though, because even though my logo is dark, I don't want the overall brand colors to skew that dark. Some of GPT's images lean too dark for my taste, but it was working from the logo and trying to figure out the visual direction from that.
Test 9: Physics atlas page with diagrams and formulas
Prompt:
Use the image generation amplifier to create a detailed physics atlas page explaining how gravity works, with scientifically accurate diagrams, formulas, annotations, force vectors, planetary orbits, spacetime curvature, mass interactions, and Einstein relativity concepts. Include labeled visual explanations, mathematical equations, cross-sections, field diagrams, and realistic textbook-style scientific illustrations. Use clean academic typography, organized educational layouts, subtle paper texture, and highly detailed information design that looks like a real modern physics encyclopedia or advanced university textbook page.Nano Banana 2:
GPT Images 2:
Verdict: Completely different styles and both are good in their own way. Nano Banana’s looks like a page from a school textbook: clean, more hand-drawn, with a few core elements explained well. GPT’s looks like a page from an actual atlas: dense, packed with information, more visual variety, and more elements on the page. Two patterns keep showing up across these comparisons: Nano Banana tends to keep things minimal while GPT fills the frame with more content, and GPT consistently produces more realistic-looking outputs while Nano Banana leans more illustrative.
Test 10: Technical infographic poster of a CPU processor
Prompt:
Create a highly detailed technical infographic poster of a futuristic CPU processor, designed like a premium engineering blueprint and semiconductor architecture sheet. Show the CPU as the central hero object with ultra-realistic metallic materials, microscopic circuitry, layered chip structures, cooling systems, thermal maps, transistor layouts, motherboard integration diagrams, nanometer process annotations, exploded internal components, and glowing data flow visualizations. Surround the processor with technical callouts, engineering measurements, scientific labels, performance graphs, manufacturing details, and advanced computing schematics. Use a dark premium background, cinematic lighting, clean scientific typography, subtle grid systems, realistic reflections, and high-end industrial design aesthetics. The composition should feel like a mix of an Intel engineering manual, NASA technical documentation, and a collectible futuristic technology poster. Ultra-detailed, photorealistic, 8K.Nano Banana 2:
GPT Images 2:
Verdict: Both are strong and this is probably the hardest one to call. They took very different creative directions from the same prompt. GPT’s looks more photorealistic and chose to use what looks like an Intel Core processor as its reference point. Nano Banana’s feels more like an illustrated blueprint. I don’t have a clear winner here. They’d serve different purposes, and both would work well as a poster or visual asset.
So which image generation model is better?
I was really curious to run this comparison because I wanted to see the patterns side by side.
And the patterns are clear. GPT Images 2 wins most rounds. It produces more realistic outputs, packs more elements into the scene, handles editorial layouts more professionally, and respects prompt intent more faithfully. When I asked for an atlas, it gave me an atlas. When I asked for a luxury brand campaign, it delivered something that looks like it came from a creative agency. Its character consistency from reference photos is on another level entirely.
Nano Banana 2 is still good, and there are situations where its cleaner, more minimal style is what you need. It’s worth knowing the differences so you can pick the right model for what you’re creating.
I mostly generate images in Claude these days using both models. Since I have both API keys set up, I switch between them depending on the result I’m going for.
You can do the same. Here's how to set up Amplifiers so you can use both Gemini and ChatGPT together inside Claude.
What do you think of the differences? Which model do you use right now?
And if you found this useful, share it with someone who’s been wondering which image generation model to use.
This post is free. Paid subscribers get access to all premium prompts and tools inside Amplifiers (the AI blew my mind MCP), weekly premium articles, all premium resources inside the AI blew my mind Lab, and exclusive partner discounts. Upgrade here.
Damn Daria this is such a good post! And has made me rethinking switching back my API to Chat...
I love this. Saving.
AI will always blow our minds.
💜ྀི🟪🫐🔮💜✨🌙💜☂️🦄💜👾🍇☂️
The magic is trying the same prompt over and over.... =)