Imagen
Imagen is Google's older line of text-to-image models, last Imagen 4 (2025). Google now calls it legacy: the Gemini API shut its Imagen models down from Aug 17, 2026 and points developers to Nano Banana instead.
Imagen is Google's family of text-to-image models, from a 2022 research project to Imagen 4 in May 2025, and as of Nov 6, 2026 it is retired for developers: Google's Gemini API documentation says "Imagen models are shut down", its deprecations table gives Aug 17, 2026 as the shutdown date for all three Imagen 4 models, and the named replacement is Nano Banana 2.1. Imagen 4 cost $0.04 an image, and $0.06 for Ultra, when it was sold in the API.
For a new project, use Google's Gemini image models, Nano Banana, which are what Google itself now recommends; Imagen matters mostly if you are moving old code off it or reading older comparisons. Outside Google, GPT Image, Midjourney, Seedream and Flux are the main rivals, compared in Best AI Image Generator.
Quick facts
| Maker | Google (Google Research, then Google DeepMind) |
| First shown | May 2022 as research, not released |
| Last version | Imagen 4, Imagen 4 Ultra and Imagen 4 Fast, released to developers Jun 24, 2025 |
| Status | Shut down in the Gemini API (Aug 17, 2026); Vertex AI advised moving off before Jun 30, 2026 |
| Replacement | Nano Banana 2.1 (gemini-nano-banana-2.1) |
| Last API price | Imagen 4 $0.04, Ultra $0.06 per image (Jun 2025) |
| Watermark | SynthID on every image |
| Open weights | No |
History
| Version | Date | What changed |
|---|---|---|
| Imagen (research) | May 2022 | A Google Research paper and demo page; Google released no code or public demo, citing the need for safeguards |
| Imagen 3 | In the Gemini API Feb 6, 2025 | imagen-3.0-generate-002; shut down Nov 10, 2025 |
| Imagen 4 | Announced May 20, 2025; API Jun 24, 2025 | Up to 2K output and much better spelling and typography, in the Gemini app, Whisk, Vertex AI and Workspace |
| Shutdown | Aug 17, 2026 | All Imagen 4 models removed from the Gemini API in favor of Nano Banana |
Sources: Imagen research page; Google I/O 2025 media models post, May 20, 2025; Gemini API deprecations, read Nov 6, 2026. Google's table notes that its shutdown dates are the earliest possible ones.
What Imagen 4 did
Text to image only. Imagen took a text prompt and returned up to four images in five aspect ratios (1:1, 3:4, 4:3, 9:16, 16:9). Imagen 4 and Ultra reached 2560 by 1792 pixels, while Fast topped out at 1408 by 768. Unlike Nano Banana, it was not a conversational model: you could not upload a photo and talk it through edits in the same way.
Typography. Google pitched Imagen 4 as "significantly better at spelling and typography" than Imagen 3, for cards, posters and comics.
Three tiers. Fast for speed and price, the standard model for most tasks, and Ultra for closer prompt following.
Provenance. Every Imagen 4 image carried an invisible SynthID watermark, and on Vertex AI also C2PA Content Credentials.
Why Google replaced it
Google's image work moved to models built into Gemini itself: Nano Banana (Aug 2025), Nano Banana Pro (Nov 2025), Nano Banana 2 (Feb 2026) and Nano Banana 2.1 (Oct 2026). These edit as well as generate, follow a conversation and draw on Gemini's world knowledge. The Imagen migration page tells developers to switch from the generate_images call to generate_content with a Nano Banana model. On the Arena text-to-image leaderboard (updated Nov 2, 2026, read Nov 6), Imagen 4 Ultra scored 1146 and Imagen 4 scored 1128, well below Nano Banana 2.1's 1325.
Imagen vs the alternatives
- Nano Banana: Google's own successor, from $0.0336 per 1K image in the API as of Nov 6, 2026, and free with limits in Gemini.
- GPT Image: OpenAI's models, which led the Arena board in Nov 2026.
- Flux: Black Forest Labs' models, with open weights for self-hosting.
- Ideogram: the pick if typography was why you used Imagen.
- Google Flow: Google's filmmaking tool, now built on Nano Banana and Veo.
Frequently asked questions
Is Imagen still available? No, not in the Gemini API: Google's documentation says Imagen models are shut down, with Aug 17, 2026 as the shutdown date for Imagen 4, and recommends Nano Banana 2.1 instead.
What replaced Google Imagen 4? Google names Nano Banana 2.1 (gemini-nano-banana-2.1), a Gemini image model released on Oct 6, 2026, as the replacement for every Imagen model.
How much did the Imagen 4 API cost? At its Jun 2025 API launch, Imagen 4 cost $0.04 per image and Imagen 4 Ultra $0.06 per image.
Was Imagen free? In 2025 Imagen 4 was part of the Gemini app, Whisk and Workspace apps under their own plans, and Google AI Studio offered limited free testing; the Imagen API itself charged per image.
Sources
- Imagen, Google AI Studio documentation, read Nov 6, 2026
- Deprecations, Google AI Studio documentation, read Nov 6, 2026
- Imagen 4 model page, Google Cloud, read Nov 6, 2026
- Fuel your creativity with new generative media models and tools, Google, May 20, 2025
- Imagen 4 now available in the Gemini API and Google AI Studio, Google for Developers, Jun 24, 2025
- Imagen, Google Research, read Nov 6, 2026
- Text-to-Image leaderboard, Arena, read Nov 6, 2026
See also
-
Nano BananaNano Banana is Google's name for its Gemini image generation and editing models. The newest, Nano Banana 2.1, came out on Oct 6, 2026; it is free to use with limits in the Gemini app and costs about $0.034 per 1K image in the Gemini API.
-
GPT ImageGPT Image is OpenAI's image generation model family, used in ChatGPT (as ChatGPT Images) and in the API. The latest, ChatGPT Images 2.6, arrived on Nov 4, 2026; it is free to try in ChatGPT with limits, and the API version costs $30 per million image output tokens. -
MidjourneyMidjourney is a paid AI image and video generator from the independent lab Midjourney, Inc. Plans run from $10 to $120 a month with no free tier, and its default model since Jul 24, 2026 is V8.2.
- SeedreamSeedream is ByteDance's family of image generation and editing models, made by its Seed team. The newest, Seedream 5.0 Pro, adds reasoning, infographics and layer separation, and costs $0.045 an image through BytePlus ModelArk.
-
FluxFlux is the family of image (and now video) models from Black Forest Labs, some with open weights you can run yourself. The newest open image model, Flux 3 Image Klein (Oct 15, 2026), is free for commercial use under $5 million in revenue, and the API starts at $0.009 an image. -
Best AI Image GeneratorThe best AI image generator in Nov 2026, by both LMArena and Artificial Analysis votes, is OpenAI's GPT Image 2.5 in ChatGPT, with Google's Nano Banana 2.1, Microsoft's MAI-Image-2.6 and SpaceXAI's Grok Imagine close behind. Adobe Firefly is the pick for commercially safe work and Flux for open weights.