Midjourney, DALL-E, Craiyon or Stable Diffusion? What were once considered strange foreign words are now a part of everyday peoples’ vocabularies as AI image generators burst in popularity.
AI art generators have quickly emerged as a shiny new tool to generate images often not possible through the human mind alone, for both skilled and the not-so-skilled creators.
What is an AI-generated image?
An AI-generated image is an image that has been created by an AI image generator, generating an image from scratch via a simple text description. The more detailed you go in your text prompt, the more insanely comprehensive your image will be.
How do AI art image generators work?
AI image generators use computer algorithms and artificial intelligence via deep learning and analysing from large datasets, creating a new image based on the prompted text.
One of the most popular methods for creating AI art includes what’s known as generative adversarial networks (GANs) consisting of two neural networks —- one which generates from the input (text), and the other that generates images from a dataset. The two networks are trained together to create images that are indistinguishable from other images, creating a feedback loop where the network continues to improve its ability to generate new images overtime.
How do I get AI generated images and what are the best AI image generators?
You can access most AI image generators for free, but some come with watermarks. Let’s start with the most popular generator: Midjourney.
Led by an independent research lab, Midjourney is a self-funded, tiny team of 11 full-time staff where images are created via text prompts in a Discord channel. You’ll need to have a Discord login to join and you can find the Discord here.
Once you’re in, you can join one of the “newbies-#” channels visible on the left sidebar. Have a peruse around the newbie channels to see how other users play with the prompts. You’ll also see people updating or upgrading their imagery, which is wild to watch, and can help give you inspiration on how to improve what you feed the model for your imagery.
In the chat function of the newbie channel, type ‘/imagine’ followed by your prompt. You can write anything that reaches your wildest imagination.
After you’ve generated an image, you can choose to either upscale or create variations of the image. You can learn more about the steps here.
OpenAI, the company behind ChatGPT, is also behind DALL-E, another artificial intelligence model that generates images directly from text. The model uses a 12-billion parameter version of ChatGPT, and combines concepts, attributes and styles.
DALL-E’s second edition goes beyond using just text prompts. Instead, it uses a technique known as ‘diffusion’, where a pattern of dots will gradually alter the pattern to resemble a picture when it recognises specific characteristics of an image. It can also create new items, because it is aware of the scene — and can add touches like lighting and shadows, as well as edit and touch-up photographs with the same description.
Are DALL-E pictures real?
DALL-E can take existing photos and upgrade them, or can create new ones from scratch.
Is DALL-E free?
DALL-E is not entirely free, as new users receive 50 free credits during their first month of use. You can access DALL-E here.
Stable Diffusion is another deep learning model created to help artists equip themselves with the tools to level-up their artistry. Users can share a text prompt and the platform will return an image that matches the text.
The platform uses a latent diffusion model, which involves adding ‘noise’ to an image, then reversing the process to improve the quality of the image. The model is used across DALL-E, Midjourney and Dream Studio. It is also free to use.
Stable Diffusion made its source code available under the Creative ML OpenRAIL-M licence, but it’s since faced controversy since making it available. Late last year, the widely popular Lensa AI app made waves across the world for allowing anyone to download the app for around $2.99, upload some selfies and then generate around 100 different avatars of themselves.
But many artists raised concerns due to the fact that the model takes site data from across the internet, so technically scrapes other people’s images in order to ‘feed’ the database. Since the Lensa App launched, many artists have called out the app for ripping off work. Since then, other companies have quickly responded that their AI image generation platforms are “ethical” and only use already-purchased images to generate new ones.
Similar to other apps, Dream by WOMBO uses two different types of neural networks — VQGAN, a network that generates images that look similar to other images, and CLIP — a neural network trained to determine whether a text description fits an image well.
You can create an image using a number of different filters, including ‘realistic’, ‘retro-futurism’ or ‘steampunk’. While images are free for everyday users, one thing holding WOMBO back is that it doesn’t allow you to create images in landscape, and is only accessible in portrait mode only.
Craiyon is a free AI image generator app, but doesn’t have the same amount of detail as some of the other models on the market. It also tends to struggle with creating ultra-realistic imagery, particularly when verbs are added into the text prompt.
Unlike applications like Midjourney, the tool is much easier to use. It also has unlimited prompts, meaning you can continue to edit the image over time, and it doesn’t cost to download your AI-generated creations.
Creepiest AI-generated images
Sometimes AI-generated images don’t always go according to plan.
In September last year, reports showed a creepy woman continued reappearing in images in which a user called “Loab” claims to have summoned while playing around with what’s known as a negative prompt. This is where you ask the model to create the opposite of what you type in. The user used a negative-weight prompt on “DIGITA PNTICS”, and the creepy woman appeared.
Earlier this year, after weird images surfaced on Stable Diffusion that still retained the Getty Images watermark, Getty Images decided to sue the creators of stable diffusion for scraping its content.
Beautiful AI imagery
Beyond the more creepy versions of AI-generated imagery, people are taking wildly epic adventures on these different AI platforms. See architects exploring AI-generated buildings here, or Design Boom’s vision for a futuristic sustainable city here.
While some people share hesitations around the speedy introduction of AI tools, others are much more welcoming. AI is only set to continue reaching into the different realms of our everyday lives. Today we have AI-generated influencers, AI models that could one day allow us to talk to animals, and even musicians like Linkin Park are now exploring the opportunities for AI video generated content.