On May 20, 2025, Google quietly flipped the switch on its most ambitious text-to-image model yet. Dubbed Imagen 4, this latest iteration promises not just breath-taking visuals but, for the first time in AI image generation, reliably legible text. For anyone who’s wrestled with garbled words on AI-generated posters or scratched their heads over mangled captions, that’s a big deal—and it’s exactly what Google is betting will set Imagen 4 apart.

AI image generators have come a long way in the past year. Recently, when OpenAI added image generation to ChatGPT, jaws dropped at the photorealism—but many users still laughed off the occasional “WELC0ME” banner or mangled comic speech bubble. Meanwhile, Google’s previous model, Imagen 3, was already winning praise for its detail, but it too struggled with typography. That all changed at Google I/O 2025, when the company unveiled Imagen 4, promising a leap forward in both image quality and text fidelity.

“Our latest Imagen model combines speed with precision to create stunning images,” explains Eli Collins, VP of product at Google DeepMind. According to Collins, Imagen 4 “has remarkable clarity in fine details like intricate fabrics, water droplets, and animal fur, and excels in both photorealistic and abstract styles.” In the sample pack, you’ll find breathtaking shots—a humpback whale breaching an icy sea, a chameleon perched on a dew-dappled leaf—that look more like high-end stock photography than computer-generated art.

Yet the real headline here is text. Google marketing materials pepper the announcement with the phrase “superior typography,” and for good reason: Imagen 4 is reportedly “significantly better at spelling and typography,” making it a breeze to craft greeting cards, posters, comics, and anything else that pairs visuals and words. In one comic strip example, every speech bubble is perfectly legible; in another, even a tiny stamped font—think passport-style microprint—is crisp enough to read without squinting.

OpenAI’s DALL·E and ChatGPT-generated images have made strides in text rendering too, but both models still occasionally drop letters or merge characters. Google’s claim is that Imagen 4 can go toe-to-toe with professional design tools when it comes to layout and legibility—an assertion that, if it holds up under real-world use, could dramatically broaden how non-designers approach visual storytelling.

If you’re itching to try it, Imagen 4 is already rolling out across Google’s ecosystem. As of May 20, you can experiment with it in:

• Gemini app (Google’s AI companion)
• Whisk (the company’s creative playground)
• Vertex AI (for enterprise users)
• Google Workspace apps like Slides, Vids, and Docs.

That means you could, say, whip up a custom slide deck in Slides with on-brand illustrations and perfectly typeset captions—all without leaving the browser. Or mock up a social media graphic in Docs, complete with headlines that don’t resemble alphabet soup.

Google isn’t stopping there. The company also teased a “fast variant” of Imagen 4, coming “soon” and capable of generating images up to ten times faster than Imagen 3. While details remain scant—Google hasn’t yet clarified whether the speed boost comes at the cost of some fidelity—the prospect of near-instant high-res image creation could be a game-changer for workflows that demand rapid iteration.

Of course, every AI model has its quirks. Google’s own documentation cautions that, despite the strides in spelling and detail, Imagen 4 may still stumble on centered compositions or extremely intricate text layouts Google DeepMind. And like all diffusion-based generators, it lacks the real-world grounding of a large language model, meaning it can hallucinate details or misplace context in complex scenes.

There’s also the question of watermarking and provenance. Google continues to embed SynthID watermarks in Imagen-generated images—a nod to ongoing industry debates over authenticity and copyright. For creators who need clean masters, that could be a minor annoyance or a legal must-have, depending on where you stand.

If Google’s claims hold up under everyday use, we could see a shift in who feels empowered to create professional-quality visuals. Small businesses might ditch stock photo subscriptions; educators could craft custom infographics on the fly; social-media managers might turn around polished memes without the Photoshop learning curve. And for the legions of casual users who love to play with AI art, the joy of seeing “Happy Birthday, Mom!” spelled correctly might just be enough to keep them coming back.