Google switches off Imagen 4 on 17 August — its image generator stops being a product and becomes a Gemini feature

Google has set 17 August 2026 as the shutdown date for three endpoints — imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001 — in the Gemini API release notes. After that date the calls simply stop working, and the migration path published by Google points developers at the Gemini image models rather than at a successor called Imagen 5. That detail is the story. Imagen was Google's answer to Midjourney and DALL-E: a named image generator with its own brand, its own model card and its own place in the product line-up. What replaces it is not another image generator but an image capability inside a general-purpose model. The same shift already happened at OpenAI, where the standalone DALL-E is scheduled to disappear from ChatGPT on 30 August. Within a fortnight, the two products that taught the public what "AI image generation" means will both have been absorbed into chat assistants. The migration is not a straight swap. Google's own guidance is inconsistent about the recommended target — some pages send developers to gemini-3.1-flash-image, others to gemini-2.5-flash-image, which is itself on a deprecation path — and the request and response shapes differ from the Imagen endpoints, so existing code has to be rewritten rather than repointed. The same release notes carry a second date that matters more to robotics than to design teams: gemini-robotics-er-1.6-preview shuts down on 31 August 2026. That was Google's publicly reachable model for spatial reasoning on robots — the one that answered questions about where objects are and how to reach them. Google has not named a replacement preview in the notes, which leaves developers building on it with three weeks' notice and nowhere obvious to go. For a catalogue like ours the pattern is worth recording: 2026 is the year the separately branded specialist model stops being how these capabilities are sold. Image generation, and possibly robot perception after it, are becoming line items in the specification of a general model rather than products with names of their own.