How to Generate AI Photos That Actually Look Real Using Nano-Banana

Nano Banana is particularly prone to this. The default output tends toward a very grey color grade that immediately signals "AI generated" to anyone who looks at it.

But there's a workaround that fixes this almost entirely. It takes about five minutes to set up, and once you understand the workflow, you can generate unlimited variations of realistic product photos, UGC-style content, and consistent character imagery without hiring a single creator.

Here's how it works.

The Core Problem (And Why Color Grading Is Everything)

The number one giveaway that an image was generated with Nano Banana is the grey colors. The tones are too neutral, the shadows lack depth, and the overall aesthetic feels clinical rather than natural.

The fix isn't to fight against these tendencies—it's to give Nano Banana a specific reference point for color and lighting that overrides those defaults.

The Pinterest-to-JSON Method

Start by finding a reference image that has the exact aesthetic you want. Pinterest is ideal for this because you can find highly curated, professionally shot images across virtually any style—warm and moody, bright and airy, editorial, lifestyle, whatever fits your brand.

Download that image. You're going to use it as the foundation for everything that follows.

Why ChatGPT Handles the Analysis

Upload your Pinterest reference image into ChatGPT 5.1 thinking mode. Ask it to analyze the photo and output a detailed JSON prompt that could theoretically recreate it.

Your prompt should be something like: "Analyze this photo and give me a very detailed JSON prompt that can recreate it. Be extremely thorough—break down the color grading, the exact colors present, the lighting direction, the shadows, the highlights, everything."

ChatGPT excels here for a few reasons. Its photo analysis capabilities are strong, and it's less lazy than Gemini when writing JSON code. You need that level of detail because the JSON is doing the heavy lifting for realistic lighting and color grading in the final image.

We're not using Claude for this step because it doesn't do actual thinking photo analysis, which is important for generating good JSON with this method.

Bringing It Into Nano Banana

Take that JSON prompt and put it into Nano Banana Pro in Google AI Studio.

Upload your product image alongside it and prompt something like: "Using this JSON as reference, generate a person holding my product."

The result should look noticeably different from a standard Nano Banana generation. The colors will have more depth, the lighting will feel more intentional, and that grey, AI-generated quality should be significantly reduced.

Building Character Consistency

Here's where it gets useful for ongoing content creation. Save that first generated character photo as your reference for future generations.

Every time you generate new content, attach that character image to maintain facial consistency. Now you have a consistent UGC model that can hold different products across multiple photos while maintaining the same face and general appearance.

This is how you build a library of UGC-style content without coordinating with actual creators. The JSON from ChatGPT handles the realistic lighting and color grading, Nano Banana handles the character consistency, and you control the product placement and creative direction.

The Iteration Loop

You can also iterate on the base JSON prompt by chatting back and forth with ChatGPT. Describe what you want, let it generate a JSON prompt based on your direction, then run that through the same Nano Banana process.

Over time, you develop a library of JSON presets that reliably produce specific aesthetics, and a roster of consistent characters that can represent your products across dozens or hundreds of images.

Five minutes to set up the initial workflow. Unlimited variations from there.

Pro Tip: For ultra realistic close-up shots of your models, add this to your prompt:

"Skin has pores, fine lines, peach fuzz, slight imperfections, oil variation. Authentic human subsurface scattering. Authentic human color complexity. Real skin stretches, compressions, and shows micro-movements. Avoid oversimplified subsurface scattering (skin looks too plastic or waxy), overly uniform texture (too smooth or too uniformly bumpy), incorrect specular response (too shiny everywhere or not shiny enough), lack of fine vascular detail, and missing the subtle translucency at edges like ears or nostrils."