Imagen 4 API Shutdown: What Broke and What to Do

The Imagen 4 API shutdown date listed in Google's own deprecation table is 17 August 2026, which was yesterday. Three model IDs are affected. If your app still calls them, the fix is not a string swap, because the recommended replacement is priced by output resolution rather than per image, and at 1K it costs more than any Imagen 4 tier did.

Here is what the primary documentation actually says, and what the new bill looks like.

The three retired Imagen 4 model IDs

From the Gemini API deprecations page:

• imagen-4.0-generate-001

• imagen-4.0-ultra-generate-001

• imagen-4.0-fast-generate-001

All three list 17 August 2026 as the shutdown date. The recommended replacement for all three is gemini-3.1-flash-image.

One detail almost every summary of this dropped, and it matters if you are triaging right now. Google's deprecation page states that the listed dates indicate "the earliest possible dates on which a model might be retired," and that exact shutdown dates are communicated to users with advance notice. So a call that still returns an image today is not evidence the date was wrong. It means your specific project has not been switched off yet. Treat that as borrowed time, not a reprieve.

What it costs now

This is the part worth doing arithmetic on before you migrate. Prices from the Gemini API pricing page, per image:

Read the two halves of that table against each other. The cheapest tier of the replacement, at 0.5K resolution, costs more than Imagen 4 Standard. At 1K it costs 3.35x what Imagen 4 Fast did. If you were generating thumbnails at the Fast tier and you migrate to 1K without thinking about it, your image bill more than triples on identical traffic.

Two levers exist and both are in the same pricing table:

• Batch pricing. Gemini 3.1 Flash Image at 1K drops from $0.067 to $0.034 in batch mode. If your generation is not user-facing and can tolerate delay, batch is roughly a halving.

• The Lite model. Gemini 3.1 Flash Lite Image is $0.0336 per 1K image, standard or batch. That is close to Imagen 4 Standard's old price and worth testing before you assume you need the full model.

A migration that will not surprise you

The mistake here is treating this as a find-and-replace on a model string. It is a resolution decision, a quality decision, and a cost decision at the same time.

1. Find every call site. One grep for imagen-4.0 across your codebase and your environment variables. Config files and background jobs are where these hide.

2. Decide the resolution before you decide the model. Look at where the output actually lands. An avatar rendered at 256 pixels does not need a 2K generation, and the pricing table charges you for the difference on every call.

3. Generate the same twenty prompts on both. Different model families interpret prompts differently. Twenty side-by-side outputs will tell you more about whether your prompt library ports than any release note.

4. Check batch eligibility per surface. User-facing generation stays synchronous. Everything else, seed data, bulk thumbnails, scheduled content, is a batch candidate at roughly half price.

5. Set a cost alert before you ship, not after. The failure mode here is silent: nothing errors, the bill just moves.

If step one turns up model names scattered across a dozen files, that is the actual problem to fix. Our walkthrough on switching AI models without breaking your app covers how to get them behind one config value so the next retirement is an afternoon instead of a sprint.

Why this keeps happening

Image models are being retired faster than text models, and the replacements are increasingly folded into general multimodal families rather than kept as standalone image endpoints. Imagen 3 was retired in November 2025. Imagen 4's preview endpoints went in February 2026. The GA endpoints went this week. That is three retirements in a family inside about ten months.

The practical read: if you build on a hosted image API, assume an eighteen-month useful life on any specific model ID and design for it. That is not cynicism, it is just the observed cadence, and it is cheaper to plan for than to discover. The broader pattern is covered in what to do when an AI model gets deprecated.

FAQ

Is Imagen 4 definitely switched off right now?

Not necessarily for your project. Google's deprecations page describes the published dates as the earliest possible retirement dates, with exact shutdown communicated to affected users. Working calls today do not mean the deprecation was cancelled.

What replaces imagen-4.0-generate-001?

Google's documentation names gemini-3.1-flash-image as the recommended replacement for all three Imagen 4 GA models.

Will my image generation cost more after migrating?

At equivalent quality, usually yes. Gemini 3.1 Flash Image starts at $0.045 for a 0.5K image against $0.02 for Imagen 4 Fast. Batch mode and the Lite variant are the two documented ways to bring that back down.

Do my Imagen prompts still work?

Treat that as untested. Prompt behaviour is model-family specific, so run a side-by-side sample before you swap the default. Prompt portability across models is a general problem, covered in writing prompts that work across AI models.

How do I catch the next deprecation earlier?

Subscribe to the changelog of every model API you call, rather than to AI news generally. That specific habit is part of a low-effort routine for tracking AI releases.