As AI technology has gotten increasingly advanced, numerous projects have emerged that attempt to generate novel images using AI methods. Such projects include OpenAI’s Dall-E 2, Google Imagen & De
ep Dream, the open-source Stable Diffusion model, and Midjourney– the main focus of this article. These models operate by receiving text that describes the desired image (the “prompt”) and outputting images tailored to the prompt. Image-generation AI models typically use a diffusion method to produce high-quality art pieces. The model first generates an image of random pixels and gradually refines the image to match the prompt.
Midjourney is an up-and-coming research lab specializing in human-AI interactions. Their main product or research project is an AI model that generates images from text, which allows users to choose images and produce variations at will. Users can gradually refine images to personal satisfaction, at which point the program can upscale the images (using AI to make the resolution of the images higher). The end results are often stunning, to the extent where Midjourney-generated images have won art competitions. In short, Midjourney is one of the most accessible, high-quality image generators available in recent years.
How do you use Midjourney? Well, generating an image is as simple as joining the Midjourney Discord server, and typing in the desired prompt that will be processed by the Midjourney Discord bot. To insert a prompt, first enter one of the text channels labeled “Newcomer Rooms.” From there, type the command “/imagine” followed by whatever comes to mind. The Midjourney Bot should deliver the desired image in 15-30 seconds, depending on the details specified.
Using the same text input, Midjourney can produce several distinct creative iterations.
Prompt: “japanese-style temple in forest turning autumn, sunset with many clouds”
Gallery: Handpicked Selections from our Midjourney Experience
Prompt: “fire and water yin yang symbol”
Prompt: “lush forests, floating island, oceans, clouds, sun in sky”
Prompt: “robot soldier, black and silver armor, running, laser sword, smoke”
Prompt: “Kids See Ghosts”
Comparison with Stable Diffusion
Prompt: “robot soldier, black and silver armor, running, laser sword, smoke”
Prompt: “lush forests, floating island, oceans, clouds, sun in sky”
Stable Diffusion is a different image generation AI, developed by a university in Munich. Unlike Midjourney, Stable Diffusion is free, but one would need to provide their own computing power to generate images.
Overall, we thought that Stable Diffusion was more photo-realistic, sacrificing the artistic flairs that the Midjourney model produced. For instance, Stable Diffusion’s floating island image didn’t convey the wonder and mystique that we thought the Midjourney variation showed. As for the robot prompt, while Stable Diffusion technically adhered to the attributes of the prompt better (actual legs and a symmetrical build), it didn’t capture the awe, intrigue, and personality of the Midjourney image.
