For a while there was this new model named nano Banana in this LLM benchmark platform LM Arena! People from AI community were speculating that it might be a totally new model from Google but turned out its just a new model in Google’s Flash model series.
nano banana is just Gemini 2.5 Flash Image Preview and was made public by Google on 26th August specifically Google highlighted that this model have amazing image editing capabilities.
so I put it to test in AI Studio and managed to get these results for image which I took at Changi Airport, Singapore on 19th August when I landed here.
I think it turned out pretty good, given I just few minutes doing the prompting while some people on X have been recommending to write long detailed prompts & that’s when the model this performs the best.
Here some prompting Nano Banana tips from Google DeepMind’s Philipp Schmid and my tip about using Gemini 2.5 Pro along with advice from Philipp to craft amazing Nano Banana images.
Also this video from Google DeepMind’s Patrick Lober covers some things about building using nano banana!
Also Josh Woodward from Google have posted this thread on X which have some cool quote tweets showing the exact the prompts that people used to generate/edit the images, if you are looking for some prompt inspirations then I would recommend checking this thread …..
And also this thread from Google Gemini App on X have some cool prompts too!
Some cool ways different companies are using nano banana to build things!
And from this whole thread I really liked this one – conversation engine to do image editing.
Memory Reconstructor uses the model to turn image generation into a continuous and conversational editing process.pic.twitter.com/umhL3gaJyS
— Google AI Developers (@googleaidevs) August 28, 2025
Also Logan Kilpatrick from Google did this podcast with Google researchers who worked on this model, I watched this earlier and it was a really interesting convo! cuz even though I’m doing SEO consulting nowadays!
But on the side I’m working towards PhD in Machine Learning too so anything ML/AI or research is something that I look forward to learning everyday.
Some main technical features of this nano banana (Gemini 2.5 Flash Image Preview mode) are ….
| Feature | Description |
|---|---|
| Reasoning + Generation | Reasons through prompt intent before generating or editing, resulting in more accurate and context-aware images. |
| Multimodal Editing | Users can generate new images, transform existing ones, combine multiple visuals, or apply style transfers using only text instructions. |
| Precision Editing | Preserves faces, key features, and stylistic details during edits, addressing a major limitation of prior generative image tools like DALL·E and Midjourney. |
Moreover this model have native world knowledge which seems contrary to GPT 5 for which it seems like OpenAI focused on Reasoning Over Memorized Knowledge as DEJAN SEO mentioned recently. But it seems like to generate better images/videos intrinsic understanding of the world and base knowledge is something that should be stored in weights.
Moreover Google recently been making major leap forwards in areas of image/video understanding and processing + how text can interact with those modalities of information.
Recent Genie 3 the 3D world model
And few weeks before that a major break through in segmentation – “Conversational image segmentation“
I think over time all these new research and technical things will help Google to improve image/video Search & make it better.
PS – I created the featured image for this blog post using this model too!
Today (31 August 2025) I’ve used Nano Banana to create this image which is related to “Web Creator confronted Google’s Danny Sullivan at WordCamp US 2025” article that I posted yesterday.
Leave a Reply