Imagen 4 API Is Dead. Here’s the Complete Migration Fix.

Google killed three Imagen 4 API endpoints on August 17, 2026, breaking production apps that call generate_images() with model IDs imagen-4.0-generate-001, imagen-4.0-fast-generate-001, or imagen-4.0-ultra-generate-001. The migration is not a drop-in swap: developers must switch to client.models.generate_content(), handle a new response structure, and loop for multiple images. Google's earlier guidance pointed to gemini-2.5-flash-image, which retires on October 2, 2026, so developers should migrate directly to gemini-3.1-flash-image (Nano Banana 2) or gemini-3.1-flash-lite-image (Nano Banana 2 Lite), both GA with no announced retirement date.

Google killed three Imagen 4 API endpoints on August 17. If your production app calls generate images with any of those model IDs, it is throwing a hard error right now — not a deprecation warning, not a graceful fallback. A hard stop. And if you migrated to gemini-2.5-flash-image based on Google’s earlier guidance, you have six weeks before that one dies too. Here is the complete fix. What Google Shut Down Three endpoints died on August 17, 2026: imagen-4.0-generate-001 Standard, $0.040/image imagen-4.0-fast-generate-001 Fast, $0.020/image imagen-4.0-ultra-generate-001 Ultra, $0.060/image Vertex AI users saw these deprecated in March. Gemini API users hit the wall yesterday. Run a codebase audit now — search for generate images , imagen-4.0 , and GenerateImagesConfig . The Migration Is Not a Drop-In Swap Three things changed at the API level. Miss any of them and you ship broken code. 1. The method is gone. client.models.generate images does not exist on Gemini image models. Switch to client.models.generate content . 2. The response structure changed. response.generated images is gone. Iterate over response.candidates 0 .content.parts and check each part for inline data . 3. Multiple images require a loop. There is no direct equivalent to number of images=4 in the same request shape. Call in a loop or check whether your target model supports the n config parameter. Here is a minimal before-and-after: Old Code Broken as of August 17 python from google import genai from google.genai import types client = genai.Client response = client.models.generate images model='imagen-4.0-generate-001', prompt='A photo of a golden retriever on a beach', config=types.GenerateImagesConfig number of images=1 , response.generated images 0 .image.save 'output.png' New Code Working with Gemini 3.1 Flash Image python from google import genai from google.genai.types import GenerateContentConfig, Modality from PIL import Image from io import BytesIO client = genai.Client response = client.models.generate content model="gemini-3.1-flash-image", contents="A photo of a golden retriever on a beach", config=GenerateContentConfig response modalities= Modality.IMAGE , for part in response.candidates 0 .content.parts: if part.inline data is not None: image = Image.open BytesIO part.inline data.data image.save "output.png" Firebase AI Logic users have a separate migration path — check Google’s Firebase migration guide https://firebase.google.com/docs/ai-logic/imagen-models-migration for the SDK-specific calls. Do Not Stop at Gemini 2.5 Flash Image Google’s initial migration guidance pointed at gemini-2.5-flash-image . That model retires on October 2, 2026 — 44 days from now. If you migrate to it today, you are doing this again in six weeks. Go directly to gemini-3.1-flash-image Nano Banana 2 or gemini-3.1-flash-lite-image Nano Banana 2 Lite . Both are GA with no announced retirement date. Google’s model retirement tracker https://vorplabs.com/models/google-model-retirements is worth bookmarking if you run multiple Google AI integrations. Which Model Is Right for Your Workload Two real options: | Model | Price 1K img | Speed | Max Res | Status | |---|---|---|---|---| gemini-3.1-flash-image | $0.067 | ~10s | 4K | GA — safe | gemini-3.1-flash-lite-image | $0.034 | ~4s | 1K | GA — safe | gemini-3.1-flash-image Batch | $0.034 | async | 4K | GA — safe | gemini-2.5-flash-image | ~$0.050 | ~7s | 2K | Retires Oct 2 | Gemini 3.1 Flash Image is the full-quality target. Four-K output, 14 aspect ratios, multi-subject consistency across up to five characters, native world knowledge. List price is approximately $0.067 per 1K-resolution image. Run it through the Batch API https://ai.google.dev/gemini-api/docs/pricing and the price drops to $0.034 with async delivery. Gemini 3.1 Flash Lite Image is the budget option. At $0.034 per image, it is cheaper than old Imagen 4 Standard $0.040 and roughly matches what Imagen 4 Fast cost in bulk. Outputs cap at 1K resolution, text rendering is weaker on small fonts, and there is no Search grounding. For draft generation, thumbnail pipelines, or high-volume preview workflows, it earns its place. At standard resolution, real-time, Nano Banana 2 costs 67% more than Imagen 4 Standard used to. That is the honest number. There is no GA equivalent to Imagen 4 Fast’s $0.020 price point. Nano Banana 2 Lite closes the gap and brings you a stable model in return. The Quick Decision Final-quality images, full resolution, or conversational image editing: use gemini-3.1-flash-image . High-volume previews, drafts, or anything where 1K is sufficient: use gemini-3.1-flash-lite-image . Need full quality at Imagen 4 Standard pricing: use Nano Banana 2 through the Batch API. Either way, migrate once to a 3.1 model and you are on GA with no announced sunset. That stability is worth the extra cost over the alternative — staying on Gemini 2.5 Flash Image and scheduling the same migration for October. For the full list of changes in the Gemini API across all model types, the official Gemini API changelog https://ai.google.dev/gemini-api/docs/changelog is the authoritative source.