The Imagen 4 API is live — but Google set a shutdown date
If you are evaluating the Imagen 4 API today, there is one fact you need before any pricing table: Google has deprecated all Imagen 4 endpoints and will shut them down on August 17, 2026. The deprecation notice is on Google’s official Imagen documentation and its model reference page, and the exact dates are tracked on the Gemini API deprecations page.
That deadline reframes the usual “which API should I use” question. Right now, most of what ranks for “imagen 4 api” falls into two buckets: Google’s own reference docs, and aggregator landing pages that sell Imagen 4 access — most of which do not mention the shutdown at all. This guide covers what the Imagen 4 API actually costs today, what aggregators charge for it, and what survives after August 17. It is written for developers deciding whether to integrate now or skip the model entirely.
One disclosure before the numbers: Modellix is an API aggregator, and this article is published on its blog. We no longer list Imagen routes in our catalog (more on why below). The pricing and dates here are sourced from Google’s own pages, accessed August 11, 2026.
The Imagen 4 family and its deprecation timeline
Imagen 4 shipped in three variants, all callable through the Gemini API under the :predict or generateImages endpoints:
| Variant | Model code | Official price (per image) |
|---|---|---|
| Imagen 4 Fast | imagen-4.0-fast-generate-001 |
$0.02 |
| Imagen 4 (Standard) | imagen-4.0-generate-001 |
$0.04 |
| Imagen 4 Ultra | imagen-4.0-ultra-generate-001 |
$0.06 |
Source: Gemini API pricing page, accessed August 11, 2026. There is no free tier for paid Imagen 4 usage; Google AI Studio offers limited free testing in previews.
Google’s deprecations page lists the full schedule. The preview endpoints (imagen-4.0-generate-preview-06-06 and imagen-4.0-ultra-generate-preview-06-06) were already shut down on February 17, 2026. The three GA endpoints shut down on August 17, 2026, and Google names gemini-3.1-flash-image (the Nano Banana family) as the recommended replacement. Imagen 3 was shut down earlier, on November 10, 2025 — this is a pattern, not a one-off.
“Imagen models are deprecated and will shut down on August 17, 2026. We recommend migrating to Nano Banana for your image generation needs.” — Google’s Imagen documentation, as displayed August 11, 2026.
Timeline compiled from Google’s deprecations page and both developers.googleblog announcements (June 24, 2025 and August 15, 2025), accessed August 11, 2026.
Why this matters for an integration decision: every line of code that calls imagen-4.0-generate-001 stops working on that date unless the route in front of it changes. That is the entire context for the pricing comparison below.
What Google charges for Imagen 4
The Gemini API pricing page lists three per-image rates: $0.02 for Imagen 4 Fast, $0.04 for Imagen 4 Standard, $0.06 for Imagen 4 Ultra. People Also Ask answers that say “Imagen 4 is $0.04 per image” are half-right — that is only the Standard tier. The model you choose changes the unit price by 3x.
The configuration surface is small. Imagen 4 accepts English-only prompts, generates 1–4 images per request, and supports imageSize (1K or 2K, Standard and Ultra only), aspectRatio (1:1, 3:4, 4:3, 9:16, 16:9), and personGeneration. Every output carries Google’s invisible SynthID watermark. None of these parameters change the per-image price.
If you are weighing imagen 4 api cost against other image models, note what the official numbers do not include: Google Cloud project setup, quota management, and the integration work that has to be redone on August 17. Direct API pricing is only part of total cost.
What aggregators charge — and what they don’t say
The middle of the SERP for “imagen 4 api” is an aggregator wall: AIMLAPI, Replicate, Vercel AI Gateway, Wavespeed, KIE, Segmind, MagicHour, and AtlasCloud all advertise Imagen 4 access. Their pages are where a developer lands when they want a key without a Google Cloud project.
The pattern across those pages is consistent: capability claims (2K output, text rendering), a sign-up button, and no shutdown mention and often no price. Replicate’s google/imagen-4 model page lists $0.04 per output image (per the live search snippet captured August 11, 2026), roughly matching Google’s Standard tier. AIMLAPI’s Imagen 4 page, as fetched the same day, shows no price at all. None of the aggregator snippets in the current SERP disclose the August 17 shutdown — which means a developer who signs up today for an Imagen 4 route is buying roughly one week of service, whether they know it or not.
This is where an aggregator’s catalog discipline shows up. Modellix’s Google provider page no longer lists any Imagen route: the Imagen model URL returns “Model Not Found”, and the Imagen series page shows zero models as of August 11, 2026. When Google announced the deprecation, we removed the routes rather than keep selling a model with a hard expiry. That is the honest version of “what happens to aggregator Imagen 4 access after August 17”: either the aggregator swaps the route to a replacement model, or the call fails.
Per-image prices from Google’s pricing page, Replicate’s model page listing (search snippet, August 11, 2026), and Modellix’s Nano Banana 2 model page, accessed August 11, 2026. Aggregator prices change; check the live model pages.
Calling Imagen 4 through an aggregator API
If you still want Imagen 4 in the remaining window, the aggregator path is a single REST call with an OpenAI-compatible shape. A generic example that works on most aggregators (Modellix included, when a route exists):
1 | curl -X POST "https://api.example-aggregator.com/v1/images/generations" \ |
The “imagen 4 api python” variant is the same call through requests or an OpenAI SDK pointed at the aggregator base URL. The response shape (base64 image bytes, sometimes with a metadata object) varies by provider, so read the aggregator’s docs for the exact field names.
The operational point of this route is not the curl call — it is that the model name lives in your request, not in your SDK. When a provider deprecates a model, the aggregator can remap that name to a successor. That is precisely what makes the August 17 migration survivable without a code change, which is the next section.
August 17: what breaks and what survives
Google’s migration guidance lists three changes for direct Gemini API users:
- Model name:
imagen-4.0-generate-001→gemini-3.1-flash-image. - Method:
client.models.generate_images(...)→client.models.generate_content(...). - Response handling: Nano Banana returns image data inside content parts, not a dedicated
generated_imagesarray.
That is a real code migration — model string, SDK method, and response parsing all change. The current Google docs name gemini-3.1-flash-image as the replacement; the docs also note that Imagen itself is deprecated and recommend moving to the Nano Banana image-generation workflow covered in the Gemini image generation guide.
An aggregator route absorbs changes 1–3 behind its own API: your request keeps the same endpoint and authentication, and the aggregator updates the model mapping (and tells you what the new model actually is — a route that silently swaps output behavior without disclosure is a liability, not a convenience). If your product already calls Google’s image models through an aggregator, the shutdown is an announcement, not an incident. For a direct integration, it is a sprint.
Our how to use Google Imagen guide walks through the direct Gemini API and Vertex AI paths in detail; the migration steps above are the summary of what changes on August 17.
How to choose: direct, aggregator, or skip Imagen
Three realistic paths, given the deadline:
- Skip Imagen 4 entirely for new work. If you are starting an integration now, you would ship on a model that dies in days. Pick the replacement first. Google’s current recommendation is
gemini-3.1-flash-image; our Nano Banana 2 API guide covers that model family’s endpoints and pricing. - Use Imagen 4 for a bounded task before August 17. Fine for one-off batches, marketing assets, or benchmark runs. Use Google AI Studio for free experiments (the “imagen 4 api free” route), and pay per image only for what you need. Just do not build a product on it.
- Migrate now and keep the aggregator abstraction. If you value one key and stable integration code across model generations, route Google image generation through an aggregator that already carries the replacement models.
There is no version of “Imagen 4 API” that is a safe long-term bet, because the model itself is being retired. The decision is between a short-lived model and its successor, not between two stable vendors.
Where Modellix routes Google image generation now
Since we removed Imagen from the catalog, Google image generation on Modellix runs on the Nano Banana family: google/nano-banana-2 (Gemini 3.1 Flash Image) is listed at $0.0403 per image at 512px and $0.0575 at 1K, with nano-banana-pro, nano-banana-2-lite, and the edit variants above and below it on the Google provider page. Pay-as-you-go, no monthly fee, and new accounts get $10–30 of free credit without a card — the same 30-minute setup as every other route.
We are not arguing that Nano Banana is “better” than Imagen 4. It is the model Google actually supports going forward, and it is the route that will still answer requests in September. If you are comparing image APIs, our cheapest AI API comparison has the honest per-image rates across providers, including where Imagen 4’s official pricing sat before the shutdown.
FAQ
How much does the Imagen 4 API cost?
Google’s pricing page lists $0.02 per image for Imagen 4 Fast, $0.04 for Imagen 4 Standard, and $0.06 for Imagen 4 Ultra, accessed August 11, 2026. There is no free tier for paid usage.
Is the Imagen 4 API shutting down?
Yes. Google has deprecated all Imagen 4 endpoints with a shutdown date of August 17, 2026, and recommends gemini-3.1-flash-image as the replacement. The schedule is on the Gemini API deprecations page.
Can I still use the Imagen 4 API for free?
Google AI Studio offers limited free testing of image models, but the paid API has no free tier. An “imagen 4 api free” search result that promises free API keys is usually an aggregator trial or a rate-limited preview.
What is the difference between Imagen 4 and Imagen 4 Ultra?
Ultra (imagen-4.0-ultra-generate-001) produces output more closely aligned with the prompt at $0.06 per image; Standard (imagen-4.0-generate-001) is $0.04. Both support 1K and 2K output. Fast (imagen-4.0-fast-generate-001) is the $0.02 speed tier.
Do aggregators still offer the Imagen 4 API?
Several do — AIMLAPI, Replicate, Wavespeed, and others list Imagen 4 routes as of August 11, 2026, mostly without mentioning the shutdown. Modellix removed its Imagen routes when Google announced the deprecation and routes Google image generation through the Nano Banana family instead.
Sources accessed August 11, 2026: Gemini API Imagen docs, Imagen 4 model reference, Gemini API deprecations, Gemini API pricing, Imagen 4 launch announcement, Imagen 4 Fast / GA announcement, Replicate model page, AIMLAPI Imagen 4 page, and Modellix’s Google provider and model pages. Google can change pricing, availability, and deprecation dates without notice. Modellix is an API aggregator and has a commercial interest in this comparison.
Cover image: illustrative Modellix artwork; it is not a Google product screenshot or source evidence.
