Google Launches Nano Banana 2 Lite and Gemini Omni Flash

Google is making two models available to developers for creating and editing visual content: Nano Banana 2 Lite, focused on generating images quickly and at low cost, and Gemini Omni Flash, designed to create and modify videos through written instructions.

Together, they support complete workflows: generate an image, use it as a reference, and turn it into a video. Google is making both available today in AI Studio, the Gemini API, and Gemini Enterprise Agent Platform.

Nano Banana 2 Lite prioritizes speed and volume

Nano Banana 2 Lite, whose technical name is gemini-3.1-flash-lite-image, is designed to generate many images in a short time. Google says it produces text-to-image results in 4 seconds and prices it at $0.034 per 1K-resolution image.

That makes it an option for prototypes, visual drafts, interactive applications, and workflows that need to generate thousands of images without sending costs soaring. For example, a store could quickly create several versions of a product photo before deciding which one to use.

Although it prioritizes speed, the model retains capabilities such as:

• Reliably following user instructions.
• Preserving characters' appearance across images.
• Creating legible text within images.

Google recommends Nano Banana 2 Lite as a replacement for its first Nano Banana model, identified as gemini-2.5-flash-image. The lineup is divided as follows:

• Nano Banana 2 Lite: speed and high-volume generation.
• Nano Banana 2: a balance of quality, speed, and cost.
• Nano Banana Pro: professional tasks that require greater control and precision.
• Nano Banana: the previous model, which Google recommends replacing.

The model is also being gradually added to consumer products such as AI Mode in Search, the Gemini app, NotebookLM, Google Photos, Stitch, Google Flow, and Google Ads.

Gemini Omni Flash brings video editing to chat

Gemini Omni Flash, available as gemini-omni-flash-preview, is reaching developers for the first time in public preview. It can accept text, images, and videos to generate clips or edit them through conversational instructions.

It costs $0.10 per second of generated video, the same price Google announces for Veo 3.1 Fast. In practice, you can request changes such as adjusting a scene, adding graphic elements, or modifying a sequence without editing every detail manually.

Its features include:

• Editing videos with natural language and holding a conversation about the changes.
• Combining text, images, and reference videos.
• Synchronizing text and graphics with actions inside the video.
• Drawing on knowledge of history, biology, or narrative logic to build scenes.

For now, each generation is limited to 10 seconds. The API also does not yet support using audio references or extending an existing scene. Although its technical schema accepts video references of up to three seconds, Google warns that they are not currently processed correctly. It also acknowledges limitations in keeping a character consistent when changing scenes or making panning movements.

The real bet is using both models together

Google's most practical proposal is to chain both systems together. First, Nano Banana 2 Lite quickly generates an image. Then, Gemini Omni Flash uses it as a reference to create an animated video.

The Interactions API can preserve a session's history and the context of its changes. According to Google, users can make up to three consecutive edits, so a request such as "make it brighter" can follow one like "move the camera closer" without starting from scratch.

Google is launching several demo applications alongside the models:

• Anywhere turns a user's selfie or uploaded photo into scenes set in well-known places and then animates the image.
• Space Lift redesigns a room from a photograph and creates a video of the result.
• Omni product studio turns product images into promotional videos for ecommerce.

For you, the impact depends on whether you use these tools or build products with them. Developers can create faster, cheaper visual generators, while users will be able to move from a written idea to an image and then to a clip in fewer steps. But the result still requires supervision, especially when the scene changes or consistent characters are needed.

Google says both models use SynthID watermarks, which let you verify AI-generated content from Gemini, Chrome, or Search. The next thing to watch is whether Omni Flash extends video length and resolves its problems with audio, references, and visual continuity.