Imagen 4 API Shut Down: How to Migrate and What It Costs

On August 17, 2026, Google switched off the last three Imagen 4 endpoints in the Gemini API. imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 no longer resolve, and Google's deprecations page points every caller at one replacement: gemini-3.1-flash-image. If a pipeline still sends requests to an Imagen model ID, it is returning errors right now. Here is what died, what replaces it, and the cheapest working path out.

TL;DR: Google retired all three Imagen 4 API endpoints on August 17, 2026. The official replacement is gemini-3.1-flash-image, which is live on Unifically as Nano Banana 2 at $0.034 per image at 1K. Imagen 4 itself was never served here; for its retired fast and ultra tiers, the closest equivalents on the catalog are Nano Banana 2 Lite at $0.02 and Nano Banana Pro at $0.07. Migration is a model-string swap plus a prompt re-test, and the $0.20 free balance on new accounts covers five test images.

Key takeaways

• Google shut down imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 in the Gemini API on August 17, 2026. The two preview variants had already gone dark on February 17.
• The migration target for all three is the same model: gemini-3.1-flash-image. The standalone Imagen line has ended, and image generation now lives inside the Gemini model family.
• gemini-3.1-flash-image is live on Unifically as Nano Banana 2, priced per image at $0.02 (Lite), $0.034 (1K), with resolution options up to 4K.
• Each retired Imagen tier has a closest equivalent on Unifically: fast maps to Nano Banana 2 Lite, standard to Nano Banana 2, ultra to Nano Banana Pro. Imagen 4 itself was never on the catalog; these are successors, not the same endpoints.
• Testing the migration costs nothing: $0.20 of free signup balance covers five Nano Banana 2 images or ten on Lite.

What Google shut down on August 17

The Gemini API deprecations page lists the schedule. All three generally available Imagen 4 variants released on June 24, 2025 reached their shutdown date together:

This was the final step in a staged retreat. The Imagen 4 preview builds (imagen-4.0-generate-preview-06-06 and its ultra sibling) shut down on February 17, 2026, and imagen-3.0-generate-002 went offline back on November 10, 2025. With the August 17 shutdown, no standalone Imagen model is callable through the Gemini API at all.

One scope note: this schedule covers the Gemini API surface (AI Studio keys, generativelanguage endpoints). Vertex AI publishes its own deprecation calendar, so enterprise deployments there should check their console rather than assume the same dates.

Why image generation moved into Gemini

Google has been collapsing its image stack into the Gemini family for a year. Imagen was a parameter-driven, standalone diffusion line: you sent a prompt, an aspect ratio, and a sample count, and got images back. The Gemini image models generate and edit in conversation, use the base model's world knowledge, and accept reference images alongside text. Maintaining both stacks stopped making sense once the Gemini line caught up on quality, and gemini-3.1-flash-image is the model Google now names as the unified replacement across the whole deprecation lifecycle.

For developers the change is bigger than a model rename. Prompting style, parameter names, and editing workflows all differ between the two stacks, which is why a migration deserves an hour of testing rather than a find-and-replace.

What breaks in your code

• Any request naming an imagen- model ID fails. Grep your codebase for imagen- before assuming you are clean; batch jobs and scheduled pipelines are where dead model IDs hide longest.
• Imagen-specific parameters do not carry over one to one. The replacement model is prompt-first, with reference images doing the work that fixed parameters used to do.
• Prompts tuned for Imagen usually still read well, but output character changes between model families. Keep a small set of representative prompts and compare outputs before switching production traffic.

Migration option 1: gemini-3.1-flash-image on Google's API

The least-change path if you are already on Google billing: swap the model string to gemini-3.1-flash-image and move to the image-output form of the generateContent call. Google's deprecations page links the current migration docs. You keep your existing key and billing account, and you inherit the conversational editing features Imagen never had.

Migration option 2: the same model on Unifically, priced per image

To be precise about what lives where: Unifically never served Imagen 4, so nothing broke here on August 17. What is live is the replacement: gemini-3.1-flash-image is the model Unifically serves as Nano Banana 2, and the wider Nano Banana family offers a closest fit for each retired tier, with per-image pricing:

The ultra mapping deserves a sentence: Imagen 4 Ultra was the quality tier, and Nano Banana Pro is the quality tier here. It runs a reasoning pass before generating, renders legible text, and holds up on photoreal materials. Our Nano Banana 2 review covers the resolution tiers in detail, and the GPT Image 2 vs Nano Banana comparison has current Arena rankings for the family.

What you gain over the retired endpoints: ten aspect ratios (1:1 through 21:9), explicit 1K, 2K, or 4K resolution selection, and up to 20 reference images per call. A working request:

curl -X POST https://api.unifically.com/v1/tasks \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "google/nano-banana-2",
    "input": {
      "prompt": "A product hero shot of a ceramic mug on a walnut desk, soft morning light",
      "aspect_ratio": "16:9",
      "resolution": "1K"
    }
  }'

Submission returns a task_id; poll GET /v1/tasks/<task_id> or set a callback_url to receive the finished image.

What migration actually costs

Billing is pay-per-use with no subscription and no credit expiry, and live rates for every variant sit on the pricing page. New accounts start with $0.20 of free balance, which is five Nano Banana 2 images at 1K, ten on Lite, or two on Pro. That is enough to run your representative prompts through the replacement before spending anything.

For scale math: regenerating a 1,000-image catalog costs $20 on Lite or $34 on Nano Banana 2 at 1K. If only the hero images need the Pro treatment, mixing tiers inside one API keeps the bill closer to the old fast-tier economics.

Frequently asked questions

Is the Imagen 4 API still available?

No. Google shut down all three generally available Imagen 4 endpoints in the Gemini API on August 17, 2026, after retiring the preview variants in February. Requests to any imagen-4.0 model ID now fail.

What replaced Imagen 4?

gemini-3.1-flash-image. Google names it as the migration target for every retired Imagen model, from Imagen 3 through Imagen 4 Ultra. The same model is served on Unifically as Nano Banana 2, priced at $0.034 per 1K image.

Is Nano Banana the same as Imagen?

No. Imagen was Google's standalone diffusion line; the Nano Banana family is Gemini-native image generation. Nano Banana 2 is gemini-3.1-flash-image, the exact model Google recommends migrating to, which makes it the practical successor rather than a lookalike.

Do Imagen 4 prompts work on the replacement?

Mostly, but re-test before moving production traffic. Imagen was parameter-driven while the Gemini image models are prompt-first with reference images, so the same words can land differently. A ten-prompt eval set catches the differences in one sitting.

Can I still use Imagen 4 anywhere?

Not through the Gemini API. Vertex AI runs its own deprecation schedule, so enterprise contracts there may differ; check the Vertex console. Self-hosting is not an option since Imagen weights were never released.

How much does the replacement cost per image?

On Unifically: $0.02 on Nano Banana 2 Lite, $0.034 on Nano Banana 2 at 1K, and $0.07 on Nano Banana Pro at 1K, billed pay-per-use. The $0.20 free signup balance covers a first test batch.

What to watch next

Three things will change this post. Vertex AI's own Imagen timeline, which enterprise callers should track separately. The Gemini image line's cadence, since gemini-3.1-flash-image will eventually get a successor of its own and this migration playbook will repeat. And Arena's image boards, where the Nano Banana family's rankings move as votes accumulate. We will update the post as each lands.

If you are mid-migration, the fastest way to know whether the replacement holds up for your workload is to run your own prompts through it: try Nano Banana 2 in the playground, free balance included.