The Nano Banana AI image trend has gone viral online. It lets people turn normal photos into hyper-realistic mini 3D figurines. However, numerous social media users are now accusing it of ignoring prompts.
The AI trend is powered by Google’s Gemini 2.5 Flash Image tool. Users can create “mini-me” versions of themselves or pets as toys. They can also create fantasy avatars with packaging that looks like real products.
The craze has spread because it’s free and needs no technical skill. Social media influencers and creators are using it across Instagram, TikTok and Twitter (now X).
However, many people have started complaining about it.
“Gave a detailed style-transfer prompt (N64/low‑poly characters, keep the sky/background, etc.), but Nano Banana makes no changes — the output looks like a lightly recompressed original. Other models (flux-1-kontext, qwen-image-edit) actually stylise correctly,” alleged a Reddit user.
“The longer I use it, the more disappointed I am. I think it may be due to Google’s strict security policy. After integrating with Gemini, nano-banana is no longer nano-banana. It has been weakened too much,” wrote another user on Reddit.
Another user commented, “I tried style transfer, and I don’t think the model can do it. You can transfer characters, though.”
“The stability of nano-banana is really a joke,” commented another.
“I get that as well sometimes. Don’t know whats the reason. I tried to generate me pixel art from image, it just returned the same image. Now I’m wondering whether it can even work with generating very specific images or its use is more like image editing (like Photoshop),” came from another.
Another user complained, “I’ve tried using a 9:16 frame as input, no luck. Tried an image, same. Tried explicitly asking for 9:16, 1080x1920px, same. It’s so frustrating that l struggle so much with dimensions. ChatGPT has the same problem.”
Some users, however, tried to help others fix the issues.
“Don’t blame the model. Your prompt just sucks,” came from one user.
“Try new page every time you try to adjust something works better, and of course it’s all about the prompting giving clear instructions is a must, imagine he is like a 6-year-old with insane Photoshop skills,” suggested another.
Google responds
Google responded to the concerns with some suggestions.
GoogleHelpCommunity advised, “To guide Gemini towards a specific tone, you should include explicit instructions within your prompt. For optimal results, provide clear and specific prompts, as detailed context leads to more accurate and relevant responses from Gemini.”
“If the issue persists, please submit feedback. On mobile, tap your profile picture or initial in the top right corner. On the web, click “Settings & help” in the bottom left corner. Kindly include #GoogleGemini in your report. Your assistance is appreciated,” it added.