When Google announced Imagen 4 at its I/O conference this past May, the company made a familiar claim: “significantly improved text rendering” over its prior release, Imagen 3. Two weeks ago, that promise moved from a keynote slide to a real-world API when Google opened Imagen 4 for paid preview in its Gemini API and for limited free trials in Google AI Studio. Alongside the baseline model priced at $0.04 per image, Google introduced a higher-precision variant, Imagen 4 Ultra, at a 50 percent premium—$0.06 per image—for use cases that demand pixel-perfect adherence to text prompts.

Google describes Imagen 4 as “your go-to for most tasks,” touting robust performance across a broad range of creative briefs. According to the official launch blog, this model strikes the balance between speed, cost, and quality, tackling everything from photorealistic landscapes to stylized illustrations with particular finesse in embedding legible, on-image text.

Enter Imagen 4 Ultra, the “precision” sibling designed to follow instructions with near-surgical accuracy. Google claims Ultra outperforms leading competitors—like OpenAI’s DALL·E 3 and Midjourney 7—in prompt fidelity, especially when prompts grow intricate. It’s the choice for brands and agencies that can’t compromise on details: menus in restaurant ads, legible signage in mockups, or even comic-book speech bubbles that must read exactly as written.

In its demo, Google showcased a gamut of test prompts, all generated with Imagen 4 Ultra:

1. Three-panel comic: A tiny spaceship named “Stardust” encounters a neon leviathan in deep space. Console alerts flash “WARNING!” while the pilot yells “CRUNCH!” and “ROOOOAAARR!”
   (Result: follows every beat, though the SFX “Had!!” typo reminded us that AI still has quirks.)
2. Vintage Kyoto postcard: “Iconic pagoda under cherry blossoms, snow-capped mountains in the distance, clear blue sky, vibrant colors.”
   (Result: technically flawless but—call it Stockholm syndrome?—we found the style a little… vanilla.)
3. Hiking couple at sunrise: “Adventurous couple waving from a rock summit, panoramic valleys below, golden light.”
   (Result: picturesque but unmistakably machine-made: soft edges, slightly unnatural skin tones.)
4. Avant-garde fashion shoot: “Model in voluminous architectural gown on alien landscape, binary sunset, surreal colors.”
   (Result: dreamy and high-concept—yet somehow, you could tell it wasn’t Vogue-level real.)

At $0.04 per image for Imagen 4 and $0.06 for Ultra, Google undercuts many competing enterprise options—though not all. Midjourney’s pay-as-you-go rate hovers around $0.05 per “fast” image for casual users, while DALL·E 3 runs roughly $0.03–$0.10 depending on resolution and volume. For agencies churning out thousands of assets, that 2-cent difference can add up.

Billing is per image generation, with no metered preview tier yet beyond the limited free AI Studio tests. Rate-limit increases are available on request, so heavy-hitters can bump up throughput ahead of product launches or ad campaigns.