Google Switches Off Three Imagen 4 Models Today
The three Imagen 4 API endpoints reach their earliest shutdown date on 17 August. The replacement is not a drop-in swap, because the method used to generate images is gone entirely.
Google has flagged 17 August 2026 as the earliest shutdown date for three Imagen 4 model IDs: imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001. Developers are directed to gemini-3.1-flash-image instead. If your code still calls those endpoints, today is the day it may stop working.
The migration is more than a find-and-replace on a model name. The generate_images() method is gone entirely, and image generation now runs through generate_content(), the same call used for text. That is a structural change, not a rename. Teams moving over need to re-test several things that are easy to assume will carry across: prompt adherence, because a different model interprets the same words differently; aspect ratios, which are specified differently; SynthID handling, SynthID being Google’s invisible watermark for AI-generated media; and then the boring but expensive pair, latency and quotas.
None of this is unusual, and that is rather the point. Google publishes a deprecations page, gives notice, and names a date. What catches people out is the gap between “we announced it” and “somebody on our team read it”. An image pipeline that has been quietly working for a year is exactly the kind of thing nobody is watching until it returns errors on a Monday morning.
There is a broader pattern here worth noticing. Google is folding its separate image product into the general Gemini model family, which is what most labs are doing: one multimodal model that handles text, images and everything else, instead of a specialist endpoint per task. That is usually good news for capability and pricing, and reliably bad news for anyone who wrote integration code against the specialist. Model APIs are not stable infrastructure in the way a database or a payment API is. They have the lifespan of a phone operating system, roughly one to two years, and planning around that is now part of the job.
What this means for you: if you have never written code against an AI API, nothing here touches you. If you have, or if a tool you rely on has, this is the reminder to go and check. Grep your projects for “imagen-4” today. Then do the thing almost nobody does: subscribe to the deprecation notices for every model API you depend on, and put a recurring reminder in the calendar to review them once a quarter. If you are choosing a provider for something you intend to run for years, ask about their deprecation policy and notice period before you ask about benchmark scores. The model that is 3 percent better is worth much less than the one that will still answer the same call in eighteen months.
Sources
Source: https://kingy.ai/ai-launch-tracker/google-will-shut-down-three-imagen-4-api-models-august-17/
AI Video Startup Higgsfield Raises 400 Million at a 5.4 Billion Valuation
DST, Goldman Sachs, Liberty Global and Intel backed the round, more than quadrupling the company's January mark. Revenue reportedly went from 200 million to 500 million annualised in six months.
