If you haven’t been living under a rock, you have probably seen images generated by Artificial Intelligence (AI) popping up everywhere. These images are usually generated with a "text to image" generation model. That means, you enter a text description ("prompt") and within a couple of minutes the model generates one or more images ("generations") depicting your prompt.
That means, you enter a text description ("prompt") and within a couple of minutes the model generates one or more images ("generations") depicting your prompt.
At the time of this writing the three most popular models are Stable Diffusion, Midjourney, and DALL·E2.
In this article, we will compare these three models in regards to:
- How to create an AI-generated image
- How much does creating AI-generated art cost?
- Can you use AI-generated images commercially?
What we will not cover is an explanation of how these AI models work under the hood. Also, we will not cover a side by side comparison of the three models. That has already been done in this great Twitter thread for example:
Let’s have a look at the three models in alphabetical order.
For the purpose of this article, we will explore the three models with the prompt "A stop-motion animation still of a cute robot standing in the forest".
Note that the time of writing (September 2022) all three models are in active development (beta phase). Therefore, this article might not represent the current state if you are reading this at a later time.
DALL·E2
DALL·E2 is developed by OpenAI and currently in beta phase.
How Do You Generate an Image with DALL·E2?
To generate an image with DALL·E2, OpenAI provides you with a minimalistic web application. However, before you can generate your first image with DALL·E2, you first have to get invited to try it out. You can apply via the waitlist [5].
Prerequisites
Invitation after application via waitlist.
However, it is not guaranteed that you will get accepted and it can take a few months to get accepted [7].
Features
In DALL·E2 you can generate images with both "text to image" and "text-guided image to image" generation anlgorithms. With the "text-guided image to image" generation algorithm, you can upload an image and erase parts of the uploaded image you would like DALL·E2 to re-generated.
You can also create variations of a generated image.
Additionally, or edit generated images with the "text-guided image to image" generation algorithm. You can even extend your generated image by generating another image on top of an already generated image with the "Generation frame" option.
Resolution and Format
In DALL·E2 all generated images have a fixed image size of 1024 by 1024.
How Much Does Using DALL·E2 Cost?
When you get access to DALL·E, you get 50 free credits that you have to use up within your first month. In every following month, 15 free credits will refill [6]. Notice, that this means the free credits won’t add up over the months.
Each prompt will create four images and it will consume roughly one credit [8].
If that is not enough for you, you can buy additional credits for 115 generations for $15 [6]. That’s about 13 cents per generation once your free credits are used up.
Can You Use Images Generated with DALL·E2 Commercially?
Yes, according to their terms [4] you can use images generated with DALL·E2 commercially:
Subject to your compliance with these terms and our Content Policy, you may use Generations for any legal purpose, including for commercial use. This means you may sell your rights to the Generations you create, incorporate them into works such as books, websites, and presentations, and otherwise commercialize them. – [4]
Midjourney
Midjourney is developed by the research lab of the same name and currently in open beta phase.
How Do You Generate an Image with Midjourney?
To generate an image with Midjourney, you can use the Midjourney Discord server.
Prerequisites
Discord account to access the Midjourney Discord server.
Features
Midjourney mainly offers "text to image" generations on their discord server. You simply send a message to the Discord bot starting with the command /imagine followed by your prompt.
Additionally, you can create additional variations of a generated image or upscale a generated image to a higher resolution.
However, you can also input URLs of one or more images for a text-guided image to image generations.
Also, you can find all your generated images on midjourney.com/app and filter and organize them there.
Resolution and Format
In Midjourney, you can create images of various formats with image resolutions of up to 2048 by 2048.
How Much Does Using Midjourney Cost?
When you first join the Midjourney Discord server, you get 25 minutes free GPU time which you can use up on your own time [3]. That is roughly equivalent to 25 free generations since one generation roughly takes one GPU minute [3]. One request will generate four options based on your prompt.
After that you can select from two plans [3]:
- The basic plan costs $10/month for 200 minutes GPU time (roughly 5 cents per generation)
- The standard plan costs $30/month for 15 hours GPU time (roughly 3 cents per generation)
If you want to use the images commercially for a company with over $1Million/year in gross revenue, you need to get the corporate plan for $600/year [3].
Can You Use Images Generated with Midjourney Commercially?
Yes, if you are a paid member, you can use images generated with Midjourney commercially according to their terms [3].
Subject to the above license, you own all Assets you create with the Services. This does not apply if you fall under the exceptions below. […]
If you are not a Paid Member, Midjourney grants you a license to the Assets under the Creative Commons Noncommercial 4.0 Attribution International License (the "Asset License"). – [3]
That also means, if you are not a paid member, you can use the images for non-commercial purposes with attribution and you can also modify them.
Stable Diffusion
Stable Diffusion was created by the researchers and engineers from CompVis, Stability AI and LAION. It is open source and was released to the public in August 2022.
How Do You Generate an Image with Stable Diffusion?
You have two options to generate images with Stable Diffusion:
- run Stable Diffusion locally on your computer or on a cloud service
- use a web application like Dream Studio
Prerequisites
If you want to run the Stable Diffusion model on your own, you will require access to a GPU with at least 10GB VRAM [2]. Huggingface provides a tutorial on how to use the Stable Diffusion model [2].
If you want to learn more, here is a great step-by-step tutorial by Jim Clyde Monge:
Run Stable Diffusion In Your Local Computer – Here’s A Step-By-Step Guide
If you don’t want to deal with any code, you can also use the Dream Studio web application. For this you will only need to register for an account.
Features
Dream Studio mainly offers "text to image" generations based on Stable Diffusion. However, it has various options like setting the number of steps to generate an image or setting the random seed.
You can also set the number of images per generation between 1 and 9.
Additionally, Stable Diffusion is able to generate images with a text-guided image-to-image generation algorithm. You can play around with it on Diffuse the Rest by HuggingFace.
Resolution and Format
With Stable Diffusion, you can create images of various formats with image resolutions of up to 1024 by 1024.
How Much Does Using Stable Diffusion Cost?
When you first register on DreamStudio beta, you get £ 2 worth of credits [1]. That is roughly equivalent to 200 free generations of one image each. The price of one image generation varies between 1 and 28 credits per image depending on the image size and how many steps you want the model to spend generating your image.
After your free trial, you can purchase additional credits in £ 10 increments [1].
Can You Use Image Generated with Stable Diffusion Commercially?
Yes, according to their terms [9], you can use images generated with Stable Diffusion commercially.
Note, that while users have forfeited copyright (and any/all intellectual property right claims) on these images, they are still public domain and can be used by anyone for any purpose, including by the user. Feel free to use images from DreamStudio Beta and the Stable Diffusion beta Discord service for anything, including commercial purposes. – [9]
Conclusion
In this article, we compared the three currently most popular image generating models Stable Diffusion, Midjourney, and DALL·E2. In this comparison we took a non-technical approach an looked at:
- how to get started with each model and how long it takes from getting started to generating your first image
- how much it costs to use each model
- whether you can use the images commercially
Below, you can find a table summarizing the comparison of the three models at the time of writing. Please keep in mind that the below table is subject to change, as the three models are in active development, and the table might already be outdated soon.
Enjoyed This Story?
Here is a collection of my other Generative AI articles:
Subscribe for free to get notified when I publish a new story.
Find me on LinkedIn, Twitter, and Kaggle!
References
[1] Dream Studio, https://beta.dreamstudio.ai/faq (accessed September 5, 2022)
[2] Huggingface, "Stable Diffusion with 🧨 diffusers", google.com. https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/stable_diffusion.ipynb#scrollTo=gd-vX3cavOCt
[3] Midjourney, "Midjourney Documentation". gitbook.io. https://midjourney.gitbook.io/docs/ (accessed September 4, 2022)
[4] OpenAI, "Terms of use". openai.com. https://labs.openai.com/policies/terms (accessed September 4, 2022)
[5] OpenAI, "DALL·E Now Available in Beta" openai.com. https://openai.com/blog/dall-e-now-available-in-beta/ (accessed September 4, 2022)
[6] J. Jang, "How DALL·E Credits Work". openai.com. https://help.openai.com/en/articles/6399305-how-dall-e-credits-work (accessed September 4, 2022)
[7] Reddit, "How much time does the DALL-E 2 waitlist take? And is it 100% sure you will get accepted if you applied?". reddit.com. https://www.reddit.com/r/OpenAI/comments/vahnqg/how_much_time_does_the_dalle_2_waitlist_take_and/ (accessed September 4, 2022)
[8] D. Schnurr, "DALL·E Editor Guide", openai.com. https://help.openai.com/en/articles/6516417-dall-e-editor-guide (accessed September 6, 2022)
[9] Stability AI, "Stable Diffusion Dream Studio beta Terms of Service". stability.ai. https://stability.ai/stablediffusion-terms-of-service (accessed September 5, 2022)
