The Water Fairy

Introduction Quite some time ago, I created a series of images centered around a little girl meeting a water fairy. They turned out okay. The one at the top of this blog was my favourite version, created by Ideogram v2.6a Turbo. Time has passed, and I do most of my AI image generation at BudgetPixel , where they have a great selection of newer models. I was interested in reviving the prompt to see how it stood up to those models. I thought I would put the results here for comparison's sake. p.s. I'm not a qualified art critic, so what you'll read are just my layman's opinions. Newer Models GPT Image 1.5 - Low, Medium, & High GPT 1.5 often tends to make very high detailed, painterly images. These three are no exceptions. Usually, I often find that there is little difference between Low , Medium , and High . That isn't the case here. Low is beautiful, but Medium is significantly better, adding much more detail to the image, like it's supposed to, and High is next level. Not only are they all detailed, but they are also very vibrant, which is one of the things I see over and over in GPT images. Seedream 4.5 Personally, I think Seedream did a wonderful job on these. The only model that made the fairies out of water. These aren't quite as bright as some of the others, but the details are there, and they're consistent. Flux 2 Flex I think this model is the nicest in the Flux 2 line (personal opinion). I'm surprised at how different the two images are, considering the Flux 2 line is supposed to be more consistent. I can get past the fairies looking different, but the radically different styles of their wings are what I'm talking about. Of course, Flux always makes beautiful, vibrant and artistic images, and these two don't disappoint. Flux 2 Max Much like with Flex, Max delivers gorgeous images. They appear a little more detailed than the ones in Flex Hunyuan Image 3 Hunyuan makes amazing images. Vibrant, detailed, and beautiful. Ideogram v3 Balanced I personally don't like this image at all. It's detailed and somewhat vibrant, but I think it did a terrible job on the fairy. She looks like one of those 2D acrylic stand images. The lily pads also look plasticy. And then there's the placement of the little girl. She's kneeling on top of the water. This is the most disappointing of all the images I made. Imagen 4 Ultra Imagen 4 usually does a pretty good job. This is a nice image, but it's not at all what I was expecting. Lucid Origin - Standard & Ultra The standard model is meant for quick iterations, and that's what it did here. The first two images show a reasonable representation of what the final image will look like, but without all the fancy details. When you look at the third image, the one made with Ultra, you see the fully fleshed out images hinted at by the first two. It's beautiful and detailed. I really like how it came out. Nano Banana I thought that Nano Banana was supposed to be known for consistency. I'm surprised that the two images are very different. I'm not mad about it. They're both great images. I do want to point out that the "little girl" is clearly not that little. These aren't my favourites, but I really like them. I love the swirly, magical lines. That first image is almost certainly going to cause this blog to be rated mature. Nano Banana Pro It's clearly Nano Banana, but on steroids. It's very beautiful (imho), and the fairy makes me think, fairy instead of pretty girl with wings . I see the same in one of the two Nano Banana images as well. Qwen The Qwen images are pretty nice. Of the Qwen variants (Z-Image Turbo and P Image), Qwen is the only one that added colourful wings. The second generation from Qwen was almost identical to the first. Z-Image Turbo As Z-Image Turbo (ZIT) is a derivative of Qwen, it's not surprising that it has a similar look. I like it. It's simple and lovely. The second image generated with ZIT included two fairies instead of one. P-Image P-Image is also a derivative of Qwen, and you can tell. This one also turned out nice, imho. The second image generated with P-Image included two fairies instead of one. Wan 2.2 Wan 2.2 was a little disappointing. The images are so inconsistent, especially by way of their concepts. That's pretty typical for Wan, though. Especially the older Wan models. I think the second image turned out quite nicely, albeit unexpectedly. Wan 2.5 Wan 2.5 definitely makes up for Wan 2.2. The images are both gorgeous and vibrant. Especially, the first one. Older Models Flux v1.1 Pro & Pro Ultra There's really no surprise here. Flux Pro and the two from Flux Pro Ultra deliver what they're known for. They're not perfect, but they're still quite striking. Both models tend to hypersexualize things, though, which is most noticeable in the third image. The prompt asks for a little girl, and she's not that. HiDream i1 Dev This image is rather flat. The wings look nice, but the image is boring overall. Closing Thoughts I love doing these model comparisons. They really help me understand the qualities each model has in various areas. To be fair, it would have been better if I made the prompt model-specific to get the full benefit of each model. Maybe next time.