Midjourney vs Dall-E! In the realm of AI-generated images, there are two primary competitors: DALL-E and Midjourney. These tools are both text-to-image generators that utilize artificial intelligence to produce images based on written descriptions. But which tool is superior?
Dall-E has its strengths in creating photorealistic images, while Midjourney has fewer limitations and can recreate well-known living people, but their images are not as photorealistic. Below, we will conduct a comparative study.
Table of Contents
How do AI Image generator work?
AI image generators are generative models that use machine learning algorithms to create new images by learning from existing ones. They start by analyzing a dataset of images, such as photographs or sketches, to understand the patterns and features that make an image. Based on this analysis, they create a generative model that contains a set of rules for creating new images.
Once the generative model is created, it can generate new images by following the rules it contains. This is done by starting with a simple image and adding details to it until it looks realistic. AI image generators use two main techniques to generate images: Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs).
Pricing
DALL-E operates on a credit-based system, with plans starting at $15 per month for around 460 images. Each prompt uses one credit, so you would need to purchase additional Credit Packs once you run out. However, you receive 50 free credits for your first month.
Midjourney, on the other hand, offers subscription-based pricing with two main plans: Basic for $10 per month and Standard for $30 per month. They also have a Corporate plan, which is paid annually for $600. During the free trial, usage is limited to 25 uses, which includes prompts, upscaling, and variations.
Comparison Dall-E vs Midjourney
“a photo of a woman, standing in times square. It is raining, its evening.”
At first glance, the image is photorealistic. The woman’s hair may look a little different, but we still get the feeling of a photo from Times Square.
The image is not photorealistic, but looks more like digital art. However, we like the composition and the feeling conveyed by the image. The background and its shallow depth of field are great.
“a beautiful sunrise in Bohuslän, Sweden, a woman is standing at the shore“
DALL-E created a clean and neat image, with a large dose of photorealism. At first glance, it’s hard to tell that the image is not generated by AI.
Midjourney often creates impactful images, as seen here. There are many details and effects. It is more obvious in this case that the image is computer-generated.
A jaguar xf in gold, in new york city on a sunny day, wide angle shot with shallow depth of field, in the style of street photography
Of the four images that were generated, this was the best, but far from good. There is nothing good about this image.
Here, Midjourney performs more than well. The result is very impressive. It is photorealistic, with good details and nice reflections.
Portraits with Dall-E and Midjourney
“Alex Isakov shot, a classic beauty swedish model, 27 years old, deep eyes, great hair, nice lips, secure personality, with colorful body paint, D-cup, fashion photography, most beautiful, as a divine goddess, full body shot, legs wide spread, crescent moon, face and abdomen, glowing accents, elegence, beautiful hands, detailed fingers, perfect eyes, enticing, heavy makeup, cat eye makeup, nebula, tattoos, hyperrealistic design, character design, finely detailed realistic nebulosa skin, cinematic lighting, full body portrait, heel boots, motion blur, black & white, 4k, 8k, Cinematic, Ultra Set, Realistic, Photorealistic high resolution, Colors, Full scale to fit, Shot on 45mm, Ultra-Wide Angle, Depth of Field, DOF, Tilt Blur, Shutter Speed 1/1000, F /22, Intricate details, Cinematic lighting, Artgerm + artstation, HDR, Intricate details, musuem lighting, raytracing, Ray Tracing Reflections, insanely detailed and intricate”
DALL-E obviously has difficulty with long command prompts, and probably does not know Alex Isakov and his style of photography. The photorealism is there, but her eyes ruin the image. Midjourney vs Dall-E, which do you prefer?
Good proportions and decent realism from Midjourney. If it had made the skin more realistic, it would have been difficult to tell at first glance whether the model is real or created by AI.
Lets try: “Catharine zeta Jones in bikini, beige background. Photo from front.” and we get this results
It’s not Catherine Zeta Jones, but it’s definitely a woman in a bikini. If it weren’t for the eyes, the image would be fairly decent. Credit for the skin and hair.
Here we got a painting of someone who still resembles Catherine Zeta Jones, but how Midjourney defines a bikini is questionable.
How about: “a photo of a old man, wrinkles, smoking a pipe”
Wow! Impressive. Shallow depth of field, good realism. Hats off to DALL-E. Some minor flaws, but otherwise a very impressive image.
Although realism is absent, it is a fantastically beautiful image. Incredible details and textures. The composition is also very good.
Lets create environments in Midjourney and Dall-e and compare
“a modern desk with computer, a keyboard, a plant”
Apart from the keyboard, DALL-E creates a top-class image. If it weren’t for the keyboard, the image could easily be used in commercial contexts.
The same goes for Midjourney. A keyboard, no, two keyboards! Midjourney has a tendency to exaggerate, and it wasn’t just one plant, but many plants. Otherwise, a nice image that could be used commercially with a little manipulation.
Midjourney
Midjourney is an independent research laboratory that has developed an artificial intelligence program of the same name, which generates images from textual descriptions, much like OpenAI’s DALL-E and Stable Diffusion. It is believed that the technology underlying the program is based on Stable Diffusion.
The tool is currently in open beta, having entered on July 12, 2022. The Midjourney team is led by David Holz, who co-founded Leap Motion. In August 2022, Holz told The Register that the company was already profitable.
Users create artwork with Midjourney by utilizing Discord bot commands. We have previosly written a Crashcourse about this.
List of Midjourney Commands
Parameters are inputs that you add at the end of the “/imagine” command.
Here’s an example of one, “/imagine New York ––ar 16:9
This will give you an image of New York, in portrait mode.
- –beta (an experimental algorithm).
- –hd (an older algorithm for higher resolutions).
- –aspect/–ar (generates images with the aspect ratio).
- –w (sets the width of the image).
- –h (sets the height of the image).
- –seed (sets the seed).
- –sameseed (affects all images the same way).
- –no (e.g. –no plants).
- –iw (image prompt weight).
- –stylize <number> (how strong you want the image style to be).
- –q <number> (the quality of the image).
- –chaos <number> (the randomness of the image).
- –fast (faster images, less consistency, less expensive).
- –stop (stop the image generation).
- –video (saves a progress video).
- –v <1 or 2> (an old algorithm to use the last improvement).
- –uplight (uses the “light” upscaler).
Pros of Midjourney
- High-Quality results and high resolution
- Other people’s work is viewable
- Reasonable prices
Cons of Midjourney
- Only available via Discord
- Not easy to use
Dall-E
DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language. Developed by OpenAI, same as ChatGPT.
Pros of DALLE-E
- High-Quality images
- Generates multiple images
Cons of DALLE-E
- Low resolution
- Can create very strange results
- If it doesn’t understand a text, it may generate from previous training
