Google Imagen (Imagen 4)
Imagen was Google DeepMind's text-to-image model line. Imagen 4, announced on May 20, 2025, added 2K output and much better spelling in images. Google has since retired Imagen from the Gemini API and Vertex AI and points developers to Nano Banana instead. Here is what Imagen did, when each version ended, and what to use today.
Using it today
How to get Imagen-style images now
Imagen is no longer offered through Google's developer APIs. These are the routes that still work.
Use Google's successor: Nano Banana
Google's docs send Imagen users to Nano Banana, the Gemini image models that generate and edit from text and photos. LazyKiwi runs Nano Banana 2 in the browser.
Migrating code from the Imagen API
Google's migration note: switch from generate_images to generate_content with a Nano Banana model, and read the image from the returned content parts instead of an image response object.
Need text inside the picture?
For posters and cards with readable words, try Nano Banana 2 or GPT Image 2 in LazyKiwi's image generator, free to start.
What it was
What Google Imagen was built for
Imagen turned a text prompt into an image. Imagen 4 was the last generation: Google sold it on fine detail, typography and speed, and shipped it in three tiers before folding image generation into the Gemini-based Nano Banana models.
Imagen 4
Fine detail and 2K output
Google's I/O post (May 20, 2025) described Imagen 4 as sharper on fabrics, water droplets and animal fur, good at both photoreal and abstract styles, and able to render up to 2K in a range of aspect ratios.
- Announced May 20, 2025 at Google I/O
- Up to 2K resolution
- Photoreal and abstract styles
Typography
Text inside images that you can read
The headline improvement was spelling: Google pitched Imagen 4 for greeting cards, posters and comics where the words in the picture have to be right.
- Much better spelling than Imagen 3
- Posters, cards, comics
- SynthID watermark on every output
Tiers
Imagen 4, Imagen 4 Ultra, Imagen 4 Fast
Standard Imagen 4 was the default; Ultra followed the prompt more closely; Fast was announced as up to 10x faster than Imagen 3. Developers got Imagen 4 and Ultra in paid preview in the Gemini API on June 24, 2025, and general availability on Vertex AI on August 14, 2025.
- Gemini API preview: June 24, 2025
- Vertex AI GA: August 14, 2025
- Up to 4 images per prompt (Vertex AI)
Retired
Shut down in 2026, replaced by Nano Banana
Google's Gemini API docs now say Imagen models are shut down and to use Nano Banana for image generation. The Gemini API retired the Imagen 4 models on August 17, 2026; Vertex AI listed June 30, 2026 as their discontinuation date.
- Gemini API shutdown: August 17, 2026
- Vertex AI discontinuation: June 30, 2026
- Google's replacement: Nano Banana
Specs
Imagen 4 specs at a glance
From Google's announcement posts, the Vertex AI model page and the Gemini API deprecation table, checked October 18, 2026.
- Developer
- Google DeepMind
- Announced
- May 20, 2025 (Google I/O)
- Variants
- Imagen 4, Imagen 4 Ultra, Imagen 4 Fast
- Inputs
- Text prompt
- Outputs
- PNG or JPEG; up to 2816x1536 (Imagen 4 and Ultra), up to 1408x768 (Fast)
- Aspect ratios
- 1:1, 3:4, 4:3, 9:16, 16:9
- Images per prompt
- Up to 4 (Vertex AI)
- Watermark
- SynthID on every output
- Status
- Shut down: Gemini API August 17, 2026; Vertex AI discontinuation June 30, 2026
- In LazyKiwi
- Not available; Google's successor Nano Banana 2 is
Versions
Imagen versions and when they ended
Release and shutdown dates from Google's Gemini API deprecation table (updated October 12, 2026).
| Model | Released | Shut down | Replaced by |
|---|---|---|---|
| Imagen 3 (imagen-3.0-generate-002) | February 6, 2025 | November 10, 2025 | Imagen 4 |
| Imagen 4 (imagen-4.0-generate-001) | June 24, 2025 | August 17, 2026 | Nano Banana |
| Imagen 4 Ultra | June 24, 2025 | August 17, 2026 | Nano Banana |
| Imagen 4 Fast | June 24, 2025 | August 17, 2026 | Nano Banana |
FAQ
Imagen FAQ
What is Google Imagen?
Imagen was Google DeepMind's family of text-to-image models. Imagen 4, announced May 20, 2025, was the final generation, with up to 2K output and much better text rendering than Imagen 3.
Is Imagen AI the same as Google Imagen?
No. "Imagen AI" is also the name of an unrelated company's AI photo-editing software for photographers. This page is about Google's Imagen text-to-image model.
Is Google Imagen still available?
Not through Google's developer platforms. The Gemini API docs say Imagen models are shut down and to use Nano Banana; the Imagen 4 models were retired there on August 17, 2026, and Vertex AI listed June 30, 2026 as their discontinuation date.
What replaced Imagen?
Nano Banana, Google's Gemini-based image generation and editing models. The Gemini API deprecation table names a Nano Banana model as the replacement for every Imagen 4 model.
When was Imagen 4 released?
Google announced Imagen 4 at I/O on May 20, 2025 in the Gemini app, Whisk, Vertex AI and Workspace. It reached the Gemini API in paid preview on June 24, 2025 and general availability on Vertex AI on August 14, 2025.
What is the difference between Imagen 4, Imagen 4 Ultra and Imagen 4 Fast?
Imagen 4 was the general model, Ultra followed prompts more closely for higher quality, and Fast traded some quality for speed (announced as up to 10x faster than Imagen 3) with a lower maximum resolution.
What resolution did Imagen 4 support?
Up to 2K. Vertex AI listed sizes up to 2816x1536 for Imagen 4 and Ultra and up to 1408x768 for Fast, in 1:1, 3:4, 4:3, 9:16 and 16:9.
Did Imagen images have a watermark?
Yes. Google said Imagen 4 outputs carry an invisible SynthID watermark that identifies them as AI-generated.
Can I use Imagen in LazyKiwi?
No. LazyKiwi offers Google's successor, Nano Banana 2, plus GPT Image 2, Seedream 5.0, Qwen Image and Z-Image in the browser with free credits on sign-up.
Imagen is retired. Its successor is one click away.
Generate and edit images with Nano Banana 2, Google's replacement for Imagen, in the browser. Free to start.
More models
Image models you can run in LazyKiwi today.