Nano Banana 2.1 is Google's updated image generation and editing model. The release may have been quiet, but the changes are worth a look: closer instruction following, better continuity across successive edits, and improved control over local changes.
Nano Banana 2.1 is already available in Neylora.
One of the most useful improvements in Nano Banana 2.1 is how it handles masks and visual annotations.
We have already added this workflow to Neylora.
You no longer have to explain in words exactly which object you want to change. Mark the relevant area, describe the edit, and send the image for generation.
For example, you can:
Showing the model where to make the change gives you more control over the edit. It is especially useful when the image already works and only one detail needs attention.
Sign in to leave a comment.
No comments yet.
A familiar problem with AI editing is that a model may make the requested change while also altering the character's face, clothes, background, lighting, or other details.
Nano Banana 2.1 brings improvements to prompt adherence and multi-turn consistency: following instructions and preserving continuity through a sequence of edits.
In practical terms, the aim is to make the requested change while keeping more of the original image intact.
Google also describes a semantic masking workflow: asking the model to change a particular element while preserving the rest of the scene. As with any AI edit, check the surrounding details rather than assuming they will remain identical.
Nano Banana 2.1 supports up to 14 reference images in a request. The model's stated reference capabilities include:
This makes it a useful option for image series, advertising, storyboards, comics, character design, and other projects in which the same people or products need to appear across different scenes.
These are model-level capabilities, not a guarantee that every detail will match. Reference limits and available controls can also depend on the interface you are using.
The update also improves image quality, including:
Another focus is product recontextualization: placing an existing product or object into a new scene with a more natural-looking result.
For example, you might start with a product photograph and place that item in a different setting while retaining its recognizable appearance.
Text rendering has been one of Nano Banana's useful strengths. Version 2.1 further improves:
For production work, still proofread every word and check the layout before publishing.
The model supports image generation at 1K, 2K and 4K resolutions.
The update also addresses artifacts that could appear in high-resolution images with unusually wide or tall aspect ratios, including 1:4, 4:1, 1:8 and 8:1.
Nano Banana 2.1 belongs to the Gemini family. Rather than treating a prompt as a bag of tags, it works with the instruction and the content of the reference images.
The model offers configurable reasoning levels:
It also supports grounding with Google Search and Google Image Search, which can be useful when working with real-world subjects and facts.
These are capabilities of the underlying model. Reasoning and search options are not necessarily exposed as selectable controls in every interface.
We have added the updated model to Neylora alongside our targeted image-editing tool.
Upload an image, mark the area you want to change, and describe the desired result.
Instead of writing:
“Change the character's jacket, but do not change their face, hair, background, pose, hands, or anything else.”
Mark the jacket and write:
“Replace it with a black leather jacket.”
The workflow feels closer to a familiar image editor: you show the model where the edit should happen instead of relying on a long description of what to leave alone.
Nano Banana 2.1 is ready to try in Neylora.
