Three Imagen 4 endpoints on the Gemini API stop answering today. Google's own model documentation states it flatly: "The Imagen 4 standard, ultra, and fast endpoints are deprecated and will be shut down on August 17, 2026." The affected model IDs are imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001.
Not a string swap
The recommended destination is Gemini 3.1 Flash Image, and the migration is a code change rather than a configuration one. Google's own migration guide is specific about what breaks: the generateImages() method on an ImagenModel instance is replaced by generateContent() on a GenerativeModel. Images no longer arrive at response.images.first(); callers must iterate the parts of the first candidate in the result object to find the image data.
The capability that quietly disappears
The sharpest change is not syntactic. Google's guide states that "Gemini Image models always return a single generated image" and advises developers to "run your generation in a loop to achieve the same result". Imagen let a caller request several candidates in one request. Anything downstream that generated four options and picked the best now issues four calls — with the latency and billing profile that implies.
What the common framing gets wrong
Two errors are circulating. The first is scope: this is described as "Google kills Imagen 4". What shuts down are the endpoints on the Gemini API. Vertex AI runs its own model lifecycle on its own schedule, and the two should not be collapsed into one obituary.
The second is firmness, and Google is the source of the confusion. The Gemini API model page states the shutdown is on 17 August. The Firebase AI Logic migration guide says all Imagen models "will shut down as early as August 17, 2026" — a floor, not a date. Trackers have picked whichever phrasing suited the headline. The stronger and more recent statement is the model page's, which is the one to plan against.
Scale unknown
How much breaks today is not knowable from outside: Google publishes no usage figures for retired endpoints, and there is no migration fee to count — pricing simply moves to standard Gemini image-generation rates. What is knowable is the direction. Google is folding a separately branded image-model line into the Gemini family, and the separate imagen- namespace on the Gemini API goes with it.