Google is shutting down the preview versions of its Gemini image generation models on June 25, 2026. If any tool or script you rely on still calls gemini-3-pro-image-preview or gemini-3.1-flash-image-preview, image generation will stop working on that date. The fix is a one-line model identifier swap to the generally available versions.
What Happened
In a May 28, 2026 changelog entry, Google announced that the two preview image models are deprecated and will be shut down on June 25, 2026. The replacements shipped the same day as stable, generally available models. Both preview endpoints stop accepting requests after the shutdown date, so any code still pointing at them will return errors rather than images.
The GA models are the same engines without the preview label. According to the Gemini API models list, gemini-3-pro-image is Nano Banana Pro, described as a professional design engine with studio-quality 4K output, and gemini-3.1-flash-image is Nano Banana 2, tuned for high-efficiency production-scale generation. We covered the wider Nano Banana Pro GA rollout when it landed.
Why It Matters for Creators
Most creators never type a model ID, but plenty depend on one indirectly. If you run a ComfyUI node, a thumbnail or batch-generation script, a Sheets add-on, or any indie app that wraps the Gemini image API, that tool has a model string baked in. When the preview endpoint goes dark on June 25, the tool breaks with no warning unless its maintainer has already updated it. Checking now takes minutes; discovering it mid-deadline does not.
Key Details
Swap the preview identifier for its GA counterpart. Nothing else in the request needs to change.
| Preview model (shuts down Jun 25, 2026) | GA replacement | Display name |
|---|---|---|
gemini-3-pro-image-preview | gemini-3-pro-image | Nano Banana Pro |
gemini-3.1-flash-image-preview | gemini-3.1-flash-image | Nano Banana 2 |
Pricing: Per the Gemini API pricing page, Nano Banana Pro standard output is about $0.134 per 1K or 2K image and $0.24 per 4K image, with batch mode roughly halving that. Nano Banana 2 standard output runs about $0.067 per 1K image.
Code change: The image generation guide shows the model passed as a plain string, for example model="gemini-3.1-flash-image", so the migration is a find-and-replace.
What to Do Next
Search your projects for the string -preview in any Gemini image call and replace it with the GA identifier from the table above. After swapping, run one test generation to confirm the output and resolution match what you expect, since the GA models are the same engines but worth a sanity check before a deadline. If you rely on a third-party tool such as a ComfyUI node or a hosted app, check that its latest version has migrated and update before June 25 rather than after. This is the same model-swap pattern we documented for the Gemini 3.1 Flash-Lite GA transition in May, and the lesson is the same: migrate while it is a one-line edit, not an outage.
