When it comes to image generation, I think Midjourney is up there with the best. Its creativity is incredible, and the way it blends visual style, mood, and composition almost feels effortless. It can transform a vague concept into something wonderfully cinematic, and its stylistic reference codes make the whole experience even richer. Each update feels like a new visual language is being invented in real time.
But even with all of this power, there’s still one area where Midjourney struggles the most: generating clean, readable text within an image. It’s a bit strange to see massive improvements in lighting, anatomy, texture and realism… while text is still this chaotic, unpredictable territory. Sometimes you have to recreate an image five, six, seven times just to get letters that don’t look like a fever dream. Worse, sometimes it approx That’s right, which somehow makes mistakes even more noticeable.
To demonstrate what I mean, I created some examples using different gestures and compared the results from both Midjourney and Google’s Nano Banana. And just to be clear before jumping in: I’m a huge Midjorin fan. I really like this tool, I use it a lot and I root for its development. But even when I talk about the text, I can’t deny it.