BLOG

Nano Banana

Another week, another AI model promising to upend the creative world. This time it's Google's "Nano Banana," now officially out in the wild as part of Gemini 2.5 Flash. The hype is focused on its ability to maintain character consistency, a notorious weakness in other models. But as a high-end retoucher, my tolerance for "good enough" is basically zero.

So, I decided to put it through a real-world test. I generated a few subjects and scenes in Midjourney, then brought them into Google's AI Studio to see if Nano Banana could composite them in a way that didn't make me want to throw my Wacom pen across the room.

My prompt was specific:
place the woman in the scene, keep original crop from scene, subject is seen from side profile, walking towards camera right, slight slow shutter speed motion blur, make sure to scale the proportions correctly, i want the subject to be on the right side of the frame.

The Good: Where the Hype is Real

lI'll give credit where it's due. The core strengths are genuinely impressive and point to a future where these tools are less of a novelty.

  • Character Consistency: This is the main selling point, and it delivers. You can genuinely take a subject, change their clothes, and drop them into different environments, and the model does a shockingly good job of remembering their face and features.

  • Seamless Compositing: The ability to blend elements from separate images—what Google calls "Photo Blending"—is powerful. For generating entourage or background elements to composite later in Photoshop, this is a legitimate time-saver.

  • Iterative Workflow: The multi-turn editing process feels intuitive. You can tell it to add a chair, then change the wall color, then adjust the lighting, and it maintains context. It’s a conversational approach to editing that, in theory, feels more natural than fiddling with layers and masks for initial mockups.

The Bad: Where It Falls Apart for Professionals

This is where the reality check comes in. For every impressive feature, there's a limitation that makes it a non-starter for finished, client-ready work.

  • Resolution: The output resolution is simply not there yet. For a workflow that often ends in high-resolution print or massive digital displays, the results are unusable without significant, and often artifact-inducing, upscaling.

  • Prompt Accuracy is a Coin Toss: For every time it nails a prompt, there are two where it ignores a key detail. It might get the scale right but forget the motion blur. It might place the subject correctly but mangle their hands. This unreliability is poison to a professional timeline.

  • Image Degradation: The more you tweak and edit in a multi-turn session, the more the image quality visibly degrades. It's like making a photocopy of a photocopy—fine details get muddy, and faces can drift into the uncanny valley.

  • The Watermark: A visible Gemini logo is automatically baked into every image. For obvious reasons, this is an immediate dealbreaker for any professional use case and requires manual removal, which negates the convenience.

  • Colorspace: Limited to sRGB

The Fine Print: Can You Actually Use This for Client Work?

This is the part everyone glosses over, but for a working freelancer, it's everything.

  • Ownership: To Google's credit, their terms are clear: you own the output. They don't claim any rights to the images you generate, which is the bare minimum for any tool hoping for professional adoption.

  • Responsibility: That ownership comes with a leash. You are fully responsible for the output. If your prompt generates something that infringes on a copyright or someone's privacy, that's on you.

  • The Copyright Trap: And here's the big one. While you can use the images commercially, you likely can't copyright them. The U.S. Copyright Office is clear that work generated solely by AI isn't protectable. This means if you create the perfect brand asset with it, a competitor could theoretically find and use the exact same image, and you'd have no legal recourse. The only way to claim authorship is through significant human modification—i.e., bringing it into Photoshop and making it your own.

  • Indemnity (for some): If you're paying for access via Google Cloud, they offer IP indemnification, meaning they'll back you up in a copyright lawsuit. It's a nice safety net for businesses, but a cold comfort for the solo freelancer trying to create unique, protectable work.

The Verdict

Nano Banana is a fascinating piece of technology and a significant step forward for generative AI. For mood boarding, rapid ideation, and creating FPO (for position only) assets, it's undeniably useful.

But is it a tool for a professional retoucher? No. Not yet.

The lack of control, inconsistent results, and low resolution place it firmly in the "tech toy" category for anyone whose job depends on precision and quality. It’s a powerful assistant for the brainstorming phase, but the moment craftsmanship and pixels matter, you’re back in Photoshop. The hype is real, but so are the limitations.