May 30, 2025 19:00:00

'GenAI Image Showdown' shows which image generation AI can generate images faithful to the prompt

Several AI companies and organizations are developing image generation AI, and each developer is promoting the high performance of their own image generation AI. ' GenAI Image Showdown ' is a website that compiles the results of inputting the same prompt into multiple image generation AIs, and you can see at a glance which image generation AI can generate an image that is faithful to the prompt.

GenAI Image Showdown

https://genai-showdown.specr.net/

Below are the results of inputting the prompt 'Two Prussian soldiers wearing spiked helmets facing each other and playing a game of throwing metal rings at each other's helmet spikes' to six types of image generation AI. The types of AI are Black Forest Labs' ' FLUX.1 [dev] ', Google's ' Gemini 2.0 Flash ', Tencent's 'Hunyuan Image 2.0', Google's ' Imagen 3 and Imagen 4 ', Midjourney's ' Midjourney V7 ', and OpenAI's ' 4o Image Generation '. Imagen 3 and Imagen 4 are grouped together because there was no significant difference between the results of the two. Of the six types of image generation AI, three types, 'FLUX.1 [dev]', 'Imagen 3 and Imagen 4', and '4o Image Generation', succeeded in generating images as prompted.

For the prompt 'Digital illustration of a star with nine points,' three apps were able to generate the correct image: 'FLUX.1 [dev],' 'Midjourney V7,' and '4o Image Generation.'

For the prompt 'A ray-traced image containing five colored cubes. The red cube is stacked on top of the blue cube. The blue cube is stacked on top of the green cube. The green cube is stacked on top of the purple cube. The purple cube is stacked on top of the yellow cube. That is, from top to bottom, the order is red, blue, green, purple, yellow. The cubes are partially translucent and made of glass,' this worked correctly for all five types except 'Midjourney V7.'

When instructed to generate a maze while also showing the correct route through the maze, only '4o Image Generation' succeeded in generating it correctly.

When the prompt was 'a 20-sided die made up of 20 prime numbers, starting with the smallest prime number,' all image generation AIs failed to generate it.

Below is a graph summarizing the results of a total of 12 tests. The most accurate answer rate was '4o Image Generation', followed by 'Imagen 3 and Imagen 4', 'FLUX.1 [dev]', 'Gemini 2.0 Flash', 'Hunyuan Image 2.0', and 'Midjourney V7'.