The ability of generative AI models like ChatGPT and Gemini to generate images with impressive quality continues to amaze me, even though I’ve seen countless examples of how good these image generation models are. These models can also edit real photos by following simple text prompts. And yes, these models can be abused to create misleading visuals that can go viral online.
The companies behind them continue to release upgraded text-to-image AI models, with Google being the latest. Google announced Imagen 4 at I/O 2025 last month, saying the new model is even faster and better than Imagen 3, which is already a high-end image generation service.
Google said at the time that Imagen 4 has “remarkable clarity in fine details like intricate fabrics, water droplets, and animal fur, and excels in both photorealistic and abstract styles.” Other improvements include 2K resolution support and improved spelling and typography.
Google also teased that a fast variant of Imagen 4 would be available soon, bringing a 10x speed improvement over Imagen 3.
Now, Google has announced that Imagen 4 and Imagen 4 Ultra are available to users, sharing a few mind-blowing examples of what the new models can do.
Cost and availability
On Tuesday, Google said in a blog post that Imagen 4 and Imagen 4 Ultra are now available to paid preview users in the Gemini API. For a limited time, Google AI Studio will also support Imagen 4 testing.
Imagen 4 will cost $0.04 per output image, one cent more than Imagen 3. Go for the Ultra version, and you’ll pay $0.06 per image. Additional tiers will be announced in the future. All Imagen 4 images will feature a non-visible SynthID watermark. However, a visible watermark won’t be present.
What is Imagen 4 Ultra?
Google describes Imagen 4 as its “best text-to-image model yet,” featuring “significantly improved text rendering” over previous models. So, what can the Ultra version do? Here’s how Google describes it:
When you need your images to precisely follow instructions, Imagen 4 Ultra is the model for you. It’s designed to produce outputs that are more highly aligned with your text prompts, achieving strong results compared to other leading image generation models.
Look at these Imagen 4 Ultra samples
Words alone aren’t enough to detail the capabilities of an advanced image generation model. Luckily for us, Google provided a few mind-blowing examples of what Imagen 4 Ultra can do.
Check out a few prompts followed by the images Imagen 4 Ultra produced below.
Prompt 1: A 3-panel cosmic epic comic. Panel 1: Tiny ‘Stardust’ in nebula; radar shows anomaly (text ‘ANOMALY DETECTED’), hull text ‘stardust’. Pilot whispers. Panel 2: Bioluminescent leviathan emerges; console red text ‘WARNING!. Panel 3: Leviathan chases ship through asteroids; console re text ‘SHIELD CRITICAL!’, screen text ‘EVADE!’. Pilot screams, SFX ‘CRUNCH!’, ‘ROOOOAAARR!’.
Prompt 2: Front of a vintage travel postcard for Kyoto: iconic pagoda under cherry blossoms, snow-capped mountains in distance, clear blue sky, vibrant colors.
Prompt 3: Photograph of an adventurous couple hiking on a mountain peak at sunrise, arms raised in triumph, epic panoramic view of valleys below, dramatic light.
Prompt 4: Avant-garde fashion editorial shot: a model in a voluminous, architectural gown standing on a shimmering, alien landscape under a binary sunset, surreal colors, high-concept, cinematic.
