Unleash your creativity – Midjourney vs. Dall-E 2
You have surely already heard about algorithms that use Artificial Intelligence (AI) for creating new images. The user enters a text prompt for the AI bot and generates several image variations using a machine-learning algorithm modelled on millions of pieces of art and text input. Do you need ideas for ancient armour design, mood for the game environment or do you just want to test multiple material variations? AI can generate a huge amount of highly detailed creative ideas in just a few minutes.
There are several models that try to push the limits of human imagination and this article is dedicated to describing the functionality of Midjourney and Dall-E 2. You will also see some of our prompts and images generated through Midjourney and Dall-E. The sky’s the limit!
Midjourney
Midjourney is still in closed beta testing and runs on a Discord bot. It is a smart choice, because many people in the gaming/art world already use this app and it is relatively easy to adapt. Behind this tool is a small team of developers. However, they bring interesting features quite often and also organize office hours on Discord once a week, where users have the opportunity to ask the developers what they are interested in or learn about future plans with this “magic tool”. Midjourney is invite-only and you will need a registered Discord account and email address to gain access.
Are you interested in what you can accomplish with Midjourney? Take a look at one of our most interesting experiments!
Prompt: Entry to the underground concrete maze, year 2040, wall overgrown with plants, epic shot, realistic, cloudy atmosphere, UE5
Parameters
There are several parameters available after entering the /imagine command. You can for example generate images with an experimental algorithm, set the aspect ratio, remove required elements from the image, set how strongly the image should be stylized, control the quality of the output, set the randomness of the image, upscalers and many others. All of these features are available in the Discord application. Remix feature that allows you to adjust the prompts as you go, so you can make small adjustments to your images if you are not satisfied with the result.
Prompt: A highly realistic scene of the futuristic SWAT Special Black Ops Soldier wearing tactical and survival gear, perspective, octane render, hyper realistic, 8k quality, denoised
Midjourney offers seamless tiling which could be very useful for material artists. Ability to use AI bot to generate variations of grunges for Bitmap to Material workflow is a very exciting prospect. It will also let you use an image as inspiration (usually with text) to guide the generation. You can even add an URL for the AI to use as a reference/inspiration.
In general, images created through Midjourney may seem more artistic and cool, similar to the work of artists on Deviant art. However, it currently has a remaster function which can create a realistic image even from a messy artistic scene, on the other hand there are limitations and it may happen that sometimes the resulting image loses its essence.
Prompt: Epic Volvo Speedster Truck, dark wine red pearlescent colors, night scenery, cinematic, hyper realistic, 8K, crystalpunk, trending on artstation
If you would like to emulate the style of different artists or anime art styles, Midjourney does this part very well. For example, a prompt in the style of Studio Ghibli or Frank Fazetta will generate remarkably convincing results, as if it were really a piece from the workshop of these authors.
Pricing
There are two types of subscriptions to choose from. Basic subscription for $10 and a standard for $30 per month. If you are a paid subscriber, you can have the bot write direct messages directly to you, instead of public rooms on discord. Images will still be visible in the web gallery but Private mode can be purchased for $20 per month. The free trial contains 25 uses which include prompts, upscaling and variations. You can experiment with Midjourney for a reasonable price if you do not mind everyone on the discord channel seeing your results.
Dall-E 2
The images API is in beta and you can get access instantly, without joining the waitlist. Dall-E 2 is Web-based and you can enjoy it on any internet browser.
Using Clip – Neutral network by open AI, Dall-E 2 can create multiple variations of outputs for your chosen prompt. Dall-E works with a diffusion model that can instantly add or remove elements from an image while being able to preserve textures, light, shadows and reflections. Dall-E can distinguish the relationship between text and image very well. The level of photorealism is high thanks to optimized sample fidelity.
Prompt: Beautiful Mercedes speedster black pearlescent paint realistic dark smoky street
Interaction with images is possible in three ways:
- Creating images from scratch using a text prompt
- Editing an existing image using a new text prompt
- Creating variations of an already existing image
Prompt: Huge colossal concrete destroyed labyrinth many walls slightly overgrown with plants, Maze runner style, realistic, atmospheric
The user has the option to edit and expand the image by uploading a mask. The image with the mask must have the same dimensions as the original image and also it has to be in png format. Transparent parts of the mask indicate places where the new content should be generated, and the prompt should contain a description for the image as a whole. Opaque parts of the masked image will not be edited.
We have used Dall-E in our Werebear project. Take a look at what this tool is capable of in the hands of skilled artists.
Images have more realism right from the first variations with Dalle-E compared to Midjourney. Image generation also takes significantly less time (approximately ten images per minute) The disadvantage may be the locked aspect ratio and also the image size of 1024×1024 px.
Pricing
Dall-E 2 works on a credit system – one prompt means one credit. You will acquire 460 images for $15 monthly. Once you run out of credits, you have to purchase another credit pack. Good news is the first month the user has 50 free credits.
Conclusion and thoughts
Both Midjourney and Dall-E are still in development and the creators are trying to improve them constantly. It is difficult to say which of these two options is better because it depends a lot on the output you expect and the impression of the generated images is quite subjective. Everyone should try both variants (or even more) and assess whether the result you get suits you and is sufficient for your needs.
Prompt: Pearlescent ambient colours generated by a invisible stain that fluoresces in ultraviolet light, highlighting livewort’s cellular structure, realistic, dynamic zoom, beautiful detail
AI will definitely be very useful for artistic inspiration and building things like textures and it will for sure shorten some parts of game/art development. This is a huge leap for everybody interested in art. You can visualise your wildest dreams in a matter of seconds! However, impressions from AI generated art are very contradictory. Many people argue that generated images lack soul, personality and are basically just pasted together pieces of pixels and images from other artists who have put time and effort into their work. On the other hand, it is tough to say which picture was created by a human being and which by a program such as Dall-E or Midjourney in some cases.
Many artists in the art world have complained that their work has been stolen and they have not received any royalties or basic credit, while anyone can now generate an image with AI within minutes to present as their own work. In art groups where people share their creations, if it is AI art, the community demands to show the progress that the person made to the image.
One thing is clear – art is only a couple of clicks away from everybody interested in it.