Gemini Image Models Hit GA: What the Flash and Pro Releases Mean for Developers
Google's gemini-3.1-flash-image and gemini-3-pro-image are now generally available, with video-to-image generation added and preview models shutting down June 25.
On May 28, 2026, Google moved its native image generation models out of preview and into general availability. If you have been building with gemini-3.1-flash-image-preview or gemini-3-pro-image-preview, there are two things you need to know: the stable versions are ready, and your preview endpoints are shutting down on June 25.
What Just Shipped
Two models are now GA:
gemini-3.1-flash-image(internally “Nano Banana 2”) — the high-volume, cost-efficient model optimised for speed and throughput.gemini-3-pro-image(internally “Nano Banana Pro”) — the quality-first model, built around advanced reasoning to handle complex instructions and render high-fidelity text in images.
Both were previously available only under their -preview suffixes. Dropping that suffix matters practically: it signals stable APIs, predictable pricing, and a production-grade SLA you can actually build on.
On the same day, Google also confirmed that gemini-3.1-flash-lite-preview has been shut down. The replacement is gemini-3.1-flash-lite.
The New Capability Worth Paying Attention To
Alongside the GA release, Google added video-to-image generation to gemini-3.1-flash-image. This is exclusive to the Flash Image model and was not available during the preview period.
The mechanic is straightforward: you pass a video file (either as a direct upload or as a public YouTube URL) as multimodal context alongside a text prompt, and the model generates a still image based on that content. The video-to-image generation guide covers the full implementation details.
The stated use cases are thumbnails, cinematic movie posters, and summary infographics. In practice, this means you can point the model at existing video content and get production-ready static assets back without manually scrubbing through footage or hiring a designer to pull keyframes. For anyone maintaining a content operation at scale, that is a meaningful time saving.
What This Means for Developers
If you are on a preview model, you have a hard deadline. Both gemini-3.1-flash-image-preview and gemini-3-pro-image-preview are deprecated and will be completely shut down on June 25, 2026. After that date, those endpoints stop responding. Migration is straightforward — drop the -preview suffix — but leave enough runway to test outputs, because GA models may behave slightly differently than what you saw in preview as Google has continued to refine them.
If you are evaluating which GA model to use, the split is fairly clean. gemini-3.1-flash-image is the right choice for high-volume generation, automated pipelines, or anywhere cost efficiency matters. It also gets you the video-to-image capability and Grounding with Google Image Search alongside Web Search. gemini-3-pro-image is the better fit for one-shot professional assets where quality and instruction-following matter more than throughput — think campaign hero images, product compositions, or anything requiring precise text rendering.
A Few Technical Details Worth Knowing
Resolution support: Both models handle 1K, 2K, and 4K output. Flash Image adds a 512px option for cases where you need smaller assets quickly.
Multi-subject consistency: Google has published a specific commitment here. gemini-3.1-flash-image maintains visual consistency across up to 5 characters and 14 objects in a single image. That number is useful if you are building anything involving repeated characters, product catalogues, or scene-based storytelling where consistency across generations matters.
SynthID watermarking: All images generated by Gemini 3.1 Flash Image carry an invisible SynthID digital watermark identifying them as AI-generated. This is applied automatically and is worth knowing if your downstream use cases involve provenance requirements or publishing guidelines.
Grounding: Flash Image can use Google Search as a tool to verify facts and ground image generation in real-time data. That opens up use cases like generating imagery tied to current events, recent products, or up-to-date reference material rather than relying solely on training data.
For Content Teams and Marketers
The video-to-image feature is the most immediately practical addition here if you are not a developer but work with one. Existing video content — product demos, YouTube series, training recordings, campaign footage — can now serve as source material for static image generation at scale. Thumbnails, poster variants, social assets, and infographic summaries can all be generated from the same video input with different text prompts.
That means less back-and-forth between creative and technical teams, and a faster path from raw video to distributable visual assets. The YouTube URL input is particularly low-friction — no file management required for publicly available content.
The Short Version
GA is live. Preview endpoints are gone on June 25. Video-to-image generation is available on Flash Image now. If you have not already updated your model identifiers, do it before the deadline. If you have not looked at what video-to-image can do for your pipeline, the documentation is a reasonable place to start.