Prompt 1: Generate a pink world where different kinds of Barbie dolls riding large sand worms from “Dune”. All sand is pink rainbow color. Sand worms have cute faces but scary teeth. Different female dolls are wearing different jobs uniforms and holding job related tools. The scene should be shot with a fish eye lens with partial motion blur effects.
Adobe Firefly Generation Above
Chat GPT4o Generation Above
Comparison summary:
Adobe Firefly has better results in fish eye effects and its Barbie is closer to the real people. However, all Barbies are wearing the same outfits that do not match the original text prompt. chatGPT 4o out performs on coherence, artistic style and characters’ outfit diversity. The flow reveals that it needs improvement on the arrangement and place of characters. The characters in the image tend to face the same direction.
Prompt 2: Generate a pink world where different kinds of female plastic dolls fighting large sand worms that is 100 times bigger from “Dune”. All sand is pink rainbow color. Sand worms have alien looking and hard scales. The scene should be pano with motion blur effects.
Adobe FireFly Generation Above
Chat GPT4o Generation Above
Comparison Summary:
ChatGPT4o outperformed on sand worms’ alien looking faces and the battle scenes. It recognizes actual rainbows instead of rainbow colors. It added weapons to the characters without asking to modify the text prompts. Adobe Firefly created much more dramatic intertwined body positions of sand worms and applied rainbow color to the sand worms themselves. However it lost the focus on Barbie dolls and made the sand worms dominant.
Prompt 3: Generate a world where different kinds of female plastic dolls wearing fancy vogue fashion clothes playing video games on first generation computers with large sand worms that is 100 times bigger from “Dune”. The video games scene is a screenshot of this generated image itself.
Adobe FireFly Generation Above
Chat GPT4o Generation Above
Comparison summary:
ChatGPT4o created much more complexed images with multiple layers and details on the dolls. However ChatGPT has stricter policies on companies’ names and copyrighted contents. I had to swamp all “Barbie” with “female dolls” to generate images. Adobe Firefly’s sand worms are separated from the dolls and computers. They rather look than interact with the dolls. Also Adobe Firefly dolls are repetitive and ChatGPT4o’ dolls are much more dynamic and detailed.
Prompt 4:Generate a world where different kinds of female plastic dolls wearing fancy vogue fashion clothes playing video games on first generation computers’ 8 bit userface with large sand worms that is 100 times bigger from “Dune”. The video games scene is a screenshot of this generated image itself. Inside the screenshot there’s another computer monitor showing the same screenshot of the screenshot. The visual should repeat 5 times or more and it get more and more glitched.
Deep Dream Generator Generation Above
Chat GPT4o Generation Above
Comparison Summary:
ChatGPT4o is the strongest AI which captures most text details, composition and meaning and translate them relatively accurate in image generation. Deep Dream AI missed sand worms completely while ignoring the original text prompts describing its 100 times bigger size. However, it does a better job on portraying Barbie dolls’ faces and computers’ closeup view. In ChatGPT 4o generated images, dolls have blurry or no faces.
Comments