Nano Banana (Gemini image)

Nano Banana behaves like a retoucher who follows instructions literally. Swap a background, restyle a subject, remove an object, keep a character identical across a series: edits that used to mean masks and layers happen by describing them, turn by turn, each request refining the last. For designers that makes it the fast surgical option between proper tool sessions, and its legible in-image text helps when a mock needs real words. It lives inside the Gemini ecosystem rather than as a standalone app, which sets the shape of its use: it is the editing brain, not the studio, so asset management, formats and brand systems stay in the design platforms that may themselves call it underneath. Outputs carry a provenance watermark, worth knowing where client expectations or disclosure norms are in play.

Example tasks

Limits

Compares

Nano Banana is the generation pick when control matters more than spectacle. Its reputation was made on instructed editing: change this, keep everything else. The same obedience shows in generation, with faithful instruction-following, legible text inside images, character consistency that holds a subject steady across a series, and quality to match. For anyone already in the Gemini ecosystem it is the ambient image capability rather than another subscription, which is the practical reason many pick it. The cinematic specialists keep the artistic ceiling, so pure spectacle briefs go elsewhere; its crown is fidelity to what you asked, not the most striking frame. Outputs carry a provenance watermark by design, a fact worth knowing where disclosure norms or client expectations apply, and the usual review applies before anything ships commercially.

Example tasks

Limits

Compares

Nano Banana set the current standard for instructed editing. Describe the change, swap this background, restyle that jacket, remove him, and it executes while preserving the rest of the scene, holding the same character identically across a whole series of edits. Photo editors spent decades approximating that capability with masks. Conversational refinement means each result takes another instruction rather than a restart. Anyone whose editing need is precise described changes without collateral damage should try it first. It is a model, not a studio: layered composition, colour-managed print work and pixel-level control remain professional-editor territory, and access runs through Google's AI surfaces rather than a dedicated editing app. Edited photographs still read as photographs, so synthetic edits of real people and events deserve honest provenance in anything published.

Example tasks

Limits

Compares