How-to·August 18, 2026, 02:00

Imagen 4 has shut down: how developers switch to Nano Banana 2

AI-generated and checked against the sources listed below.

Google has shut down its Imagen image generation models and is pointing developers to the Nano Banana models. The switch is mostly about swapping the model name and testing your prompts again.

AI-generated image

If you or your workplace have an app that creates images with Google's Imagen 4, the code needs to be updated. According to Google's Firebase documentation, all Imagen models shut down on August 17, 2026. Google recommends switching to the Nano Banana models.

Nano Banana is the popular name for Google's Gemini image models. Nano Banana 2 is technically called gemini-3.1-flash-image. That is the model Google points to as the replacement for most Imagen 4 variants.

Which model replaces what?

According to Google's Firebase guide, the replacements are:

- imagen-4.0-fast-generate-001 should be replaced by gemini-3.1-flash-image with minimal "thinking", meaning a little thinking time. - imagen-4.0-generate-001 should be replaced by gemini-3.1-flash-image with high thinking. - imagen-4.0-ultra-generate-001 should be replaced by gemini-3-pro-image.

So the regular and the fast Imagen models get the same replacement. The difference lies in how much the model is allowed to think before it draws.

What does the developer need to change?

For many, it is in practice a change of model name. But in Firebase you also need to do a bit more:

- Swap ImagenModel for GenerativeModel in the code. - Set the response type to IMAGE in the settings. - Use SafetySetting instead of ImagenSafetySettings. - Move settings from ImagenGenerationConfig to GenerationConfig.

A third-party guide recommends saving a small set of typical prompts and comparing the results before moving real traffic over. The images may look different with a new model.

Differences to know

Google points out that the Gemini image models always deliver PNG files with an invisible SynthID watermark that marks the image as AI-generated. They create one image at a time and do not support negative prompts, where you describe what should not be included. Control over how people are generated is not available either.

If you have built a solution that requests many images at once, you will need to make multiple calls.

What does it mean for you?

As an ordinary user, you will not notice anything directly. But if a tool you know uses Imagen in the background, the style of the images may change. Developers should use the exact model names in code, logs and billing rather than the nickname Nano Banana.

Sources

More on this topic

Get the week's AI news in your inbox

Choose your level, topics and length. One email a week, unsubscribe at any time.

Subscribe to Promptly Newsletter
PromptlyNewsletterRSSLog in

The news on aijour is AI-generated and checked against the cited sources.