In this blog post, I will compare how different generative AI models respond to the same prompts by analyzing the images they produce.
Prompt: Picture of a Brown Dog Eating Ice Cream
Left: FLUX.1
Right: Stable Diffusion 3
Both images clearly showcase a brown dog eating ice cream. The image on the left includes a human hand and has a more detailed, realistic background. This makes the image more natural. The dog is lighter and eating a more extravagant ice cream choice. In contrast, the image on the right focuses solely on the dog with a simpler, more stock-like background. The dog is darker, eating a more simple choice which makes the image feel clean but less detailed.
Prompt: An Elephant Riding a Skateboard
Left: SSD-1B
Right: Kadinsky 2.2
The image on the left of the elephant on the skateboard appears more realistic, with high quality detail and vibrant coloring. This gives the image a lifelike appearance. In contrast, the image on the right is more cartoon-like. It depicts the elephant riding a skateboard in a sketched form with more muted colors. While both images follow the prompt accurately, they differ significantly in style and level of realism.
Prompt: A Bird Wearing a Birthday Hat
Left: Kandinsky 2.2
Right: FLUX.1
Both images depicting a bird wearing a birthday hat are very similar and have little variation compared to the previous examples. Both birds are colorful, facing in the same direction, and have a similar realistic looking style and shape. The main differences can be seen in the environment and framing. The image on the left shows the bird indoors and includes the bird’s full body, whereas the image on the right is set outdoors and only shows the upper half of the bird. Besides the differences in color palette, environment, and size, the two images are consistent in appearance.