Google's Imagen 4: A Step Up or Just More of the Same?

Google has rolled out Imagen 4, the latest in its line of text-to-image models. This new version claims to be a significant upgrade over its predecessor, Imagen 3. For those willing to spend more, there's also Imagen 4 Ultra, which promises even more precise image generation based on text prompts. Both models are now available for testing, either through a paid preview in the Gemini API or a limited free trial in Google AI Studio. The standard Imagen 4 model is marketed as a versatile tool for most tasks, priced at $0. 04 per image. Meanwhile, Imagen 4 Ultra is aimed at users who need highly accurate results, with a 50% price hike to $0. 06 per image. Google showcased several examples, including a comic strip featuring a spaceship under attack by a giant blue lizard, complete with sound effects. The images were technically impressive, but they lacked the charm and creativity that human artists bring to their work.

Another example was a vintage travel postcard of Kyoto, complete with cherry blossoms, snow-capped mountains, and a clear blue sky. While the image was technically accurate, it felt generic and lacked any unique artistic flair. Other examples included a couple hiking and an avant-garde fashion shoot. All the images were high-quality and followed the text prompts precisely, but they still had that unmistakable machine-generated look. So, is Imagen 4 a game-changer? It's a step up from previous versions, but it's hard to say if it's a significant improvement over other leading models like Dall-E 3 and Midjourney 7. The public's initial excitement about AI-generated art seems to be waning, with many people now associating it more with spammy ads than with genuine creativity. While Imagen 4 is a solid tool, it's not yet clear if it can truly compete with the best in the market.

Actions