In this video I tested Qwen-Image (yeah… the naming is a mess—Gwen, Qwen 2.5, Qwen 3, Max… it’s hard to keep up). But the model itself is genuinely interesting because it has a big focus on complex text rendering and precise image editing. The main thing I kept watching for is whether the text actually feels like it belongs in the image—like it was painted by the same “artist”—instead of looking pasted on later like you sometimes get with Midjourney or Sora. And from the samples and my own tests, Qwen-Image does a really solid job blending text into materials like shields, banners, cardboard signs, and even sand writing. I ran a bunch of prompts through Qwen Chat and compared different styles: realistic, anime, illustration-heavy, and poster-like setups. My favorite results were honestly the illustrative/anime outputs—the armor textures and overall detail looked beautiful, and the anime conversion came out surprisingly high quality. On the realism side, it can still feel a bit “uncanny” (some subjects looked rubbery, like a 3D game model), and I did see missing letters in a couple text tests. Overall, I’d put it top 3 right now for text-in-image, and around top 4 for general image generation—especially if you run it locally with more steps (30–50) and upscale.