Introducing Gemini 2.5 Flash Image Edit aka "nano-banana"

Team fal

• 3 min read
Introducing Gemini 2.5 Flash Image Edit aka "nano-banana"

For weeks now a secret image editing model has been making waves on LMArena. The consensus speculation has been Google was behind the so called "nano-banana" and today this speculation has been confirmed. nano-banana is Google's Gemini 2.5 Flash Image editing model and we are proud to announce it is publicly available on fal.

Flexible and Powerful

nano-banana is an incredibly powerful state-of-the-art image editing model.

It understands a wide variety of natural language image editing instructions.

Input (left) and output (right) with the prompt "restore this image"

Not only can nano-banana perform precise edits, but it can also change the composition entirely, while preserve the identity of the subjects.

Input (left), output (right) with prompt "This woman youtube thumbnail excited pose extremely big colorful font "Wait nano-banana is on fal?"

Consistent Scenes

Because nano-banana is so good at preserving characters identity, it can be used to create storyboards, while preserving the characters and locations of the original image.

Four images created in sequence using compositions instructions, but preserving the identity of the character.

nano-banana perform a variety of camera motions, such as turning and zooming out.

For the first edit, we create a real world version. The next edits show nana-banana ability to reposition characters while preserving their identity.

Being able to keep characters consistent unlocks use cases such as consistent character sheets.

We can create a character sheet with the prompt "create a character turnaround with front, side, and back views.The image on the left will not be changed"

Multi-image Editing

nano-banana can not only edit a single image, but it can combine multiple images together in complex ways according to the prompt.

Multi-image editing is particular useful for virtual product photography.

The inputs (left and center) are combined to the produce the output (right) with the prompt "This man holding this bottle and advertising it, different pose, no water around it"

Multi-image can also be used with virtual try-on with no additional training.

The inputs (left and center) are combined to the produce the output (right) with the prompt "This woman wearing this bag and advertising it, different pose"

nano-banana is able to understand complex high level instructions, allowing it to turn ingredients into a final dish.

the inputs (left) are combined to produce the output (right) using the prompt "combine the ingredient and make a quiche on a plate"

Zero-shot Portrait Generation

nano-banana does a great job creating portraits from single photos, but when using multiple photos you get enhanced identity even with dramatic restyling.

Using the seven images (left) a variety of stylized portraits are possible with different prompts.

So Much Potential, So Little Time

nano-banana is an addictive model. The image quality and edit ability is unmatched. We can't wait to see what are users are able to build with it.

Try it today on fal. Go bananas: https://fal.ai/models/fal-ai/gemini-25-flash-image/edit