Imagen 3 is an advanced text-to-image model developed by Google. It’s capable of generating realistic and detailed images based on natural language prompts.
You can access the model’s capabilities for free through two web-based tools: ImageFX, an experimental platform hosted by Google Labs, and Gemini – Google’s AI chatbot and chief rival to ChatGPT.
Trained on detailed images, Imagen 3 is highly regarded for its ability to produce realistic compositions, as well as its accurate understanding and execution of different visual styles. Want to learn more and use it for yourself? We’ve broken down all the basics below.
This article was correct as of February 2025. AI tools are updated regularly and it is possible that some features have changed since this article was written. Some features may also only be available in certain countries.
What is Imagen 3?
Imagen 3 is an AI-powered text-to-image model developed by Google DeepMind, the company’s AI research lab. First announced at Google I/O in May 2024, access to Imagen 3 was opened up in August. Like other AI image generators, the model allows users to create images in a range of visual styles based on simple, natural language prompts.
Imagen 3 builds on previous versions of the technology. According to Google, the third iteration generates images with “better detail, richer lighting and fewer distracting artifacts.” The tool also allows users to refine the resulting images, editing text prompts to add specific details.
Imagen 3 has been integrated with Gemini, Google’s rival to ChatGPT, allowing users to access the model’s capabilities as part of conversations with the AI-powered chatbot.
What can you use Imagen 3 for?
Imagen 3 can generate detailed images in a matter of seconds, based on text-based descriptions. Trained on millions of images, it excels at replicating different visual styles. Prompts such as “cinematic”, “surreal” and “35mm film” can be used to produce images with a particular aesthetic. It’s particularly effective at replicating photorealism.
Accessed through ImageFX, Imagen 3 generates images in sets of four, while prompts in Gemini create one image at a time. Images can be individually copied and downloaded, while the prompt itself can be refined with specific details or art styles to fine-tune the results.
Five aspect ratios are available: square (1:1), portrait (9:16), landscape (16:9), mobile portrait (3:4) and mobile landscape (4:3).
Users own the copyright in all images produced by Imagen 3, meaning it’s a potentially useful tool for everything from editorial to product marketing.
What can’t you use Imagen 3 for?
Imagen 3 is only capable of generating still images. DeepMind is developing a separate AI-powered text-to-video generator called Veo 2.
Imagen 3 can’t be used to generate high-resolution images. Native output resolutions are limited to 1024×1024 for square images, 1408×768 for 16:9 images and 1280×896 for 4:3 images.
The model is also subject to a number of ethical policy restrictions. You can’t use it to create images of real people, for example, or to generate visuals which are potentially harmful or offensive. Imagen 3 is also trained to avoid generating images which could infringe on copyright.
While Imagen 3 is good at following detailed prompts, it doesn’t support precise editing of specific elements within an image.
How much does Imagen 3 cost?
Imagen 3 is free to use through ImageFX or Gemini. What’s more, you own the copyright in images created with Imagen 3 and can use them for any purpose, without paying royalties.
There is one limitation: if you want to generate images featuring people, you’ll need to take out a Gemini Advanced subscription. This costs $19.99 / £18.99 / AU$30 per month. The plan also includes priority access to new features and access to Google’s latest models.
Where can you use Imagen 3?
Imagen 3 can be accessed through Gemini, Google’s answer to ChatGPT. Request an image in your conversation with the chatbot and it will generate one asset at a time. You can also edit your original prompt to refine the results. Gemini can be used via the web app, as well as the app for iOS and Android.
Imagen 3 is also available through ImageFX, a dedicated AI image generator hosted by Google Labs. ImageFX benefits from an intuitive web-based interface, with suggested prompt structures and art styles. Usefully, it generates four images at a time from a single prompt.
Is Imagen 3 any good?
From our hands-on experience with both ImageFX and Gemini, we know that it’s capable of producing rich, dynamic images which broadly follow the theme of the prompt.
Online reviews have commented to similar effect, noting the high quality of results on the whole, as well as how effectively Imagen 3 handles different art styles. Our senior AI Editor, Graham Barlow said, “Imagen 3 is going to deserve a place amongst the best AI image generators.” Photorealism is a point on which it scores highly, particularly how it’s able to replicate depth of field effects and cinematic moods.
The prompt-based interface is pretty foolproof, although the model doesn’t always follow detailed edits to the letter. While it generally observes the essence of prompts, it isn’t immune from the occasional hiccup when interpreting specific details.
Use Imagen 3 if…
You want a free AI image generator
Imagen 3 is available for free through ImageFX and Gemini, allowing you to generate realistic images from text descriptions with few limitations. You’ll only need to pay if you want to generate images featuring people.
You want to experiment with art styles
Imagen 3 excels when it comes to replicating different visual styles, whether that’s cinematic photorealism or surreal illustrations. By changing a few words in a prompt, you can give your image an entirely different aesthetic.
Don’t use Imagen 3 if…
You want granular editing options
Imagen 3 allows you to fine-tune images by changing the wording of your text prompt, but you can’t selectively edit specific details or sections of an image. For this, you’ll need an alternative tool such as DALL-E 3.
You need total attention to detail
Google’s text-to-image model does a pretty decent job of following prompts. However, its interpretation of nuanced descriptions can often be a little loose. Changes to one part of a prompt can often throw off the detail in another.
Also consider
- DALL-E 3 is a text-to-image model developed by OpenAI. It’s accessible through ChatGPT and Microsoft Designer. It’s easy to use and effective at following complex prompts. However, it doesn’t score as highly as Imagen 3 for realism.
- Midjourney is an AI image generator rooted in a Discord community. There’s a bit of a learning curve with its interface, but the tool is capable of creating truly artistic images. It also features powerful editing tools and lends itself to inspiration.