The year 2022 may go down in history as the year AI art became popular.
A flood of high-quality tools from many sources, based on various AI models, is making AI art available to anybody with a smartphone and an Internet connection. An AI model is used by the tools to turn text input, known as a prompt, into an image.
The prompt is critical: Changing a single word may have radically different consequences. “Prompt engineering” is rapidly becoming a valued talent, because models trained on the same data with the same prompt should generate the same results. There’s even a burgeoning industry for prompts that produce specified outcomes.
Here are five resources to get you started. I gave them everyone the identical stimulus to compare: “A human and a robot standing near a great oak tree on a hill with clouds in the sky.”
DALL-E 2
OpenAI, which was created in 2015, made news in 2020 with the release of GPT-3, a natural-language model. In January 2021, the DALL-E digital image model was released, which has now developed into DALL-E 2. The OpenAI model produces high-quality photos in a broad range of styles. Particular prompts may result in specific results, or you might provide a broad suggestion and enjoy multiple wildly varied outcomes.
DALL-E 2, which is currently available to everyone through OpenAI’s website, is the greatest tool for people who want to know what all the fuss is about. It’s fast, outperforming others I’ve tested by a wide margin, and the website is simple to use. It returns four answers at once, usually in quite varied styles, reducing the number of times you need to perform a prompt. The findings of DALL-E 2 are also positive. It is the only AI model that shows both the human and the robot.
Stable Diffusion / Dream Studio
Stable Diffusion from Stability AI is popular for the same reasons that DALL-E 2 is: it’s fast, effective, and can generate passable pictures from a broad range of cues.
Anyone may use Stable Diffusion for free by visiting the Stable Diffusion demo website. It is not as fast as DALL-E 2, but it normally provides findings in 30 seconds or less. It also offers four versions at the same time, similar to DALL-E 2.
Because the Stable Diffusion model is open source, serious users may fine-tune how it operates. This has increased its appeal as fans flock to the model. Dream Studio, a commercial program developed by Stability AI, is based on Stable Diffusion. It offers a free trial before selling credits to produce new photos. In exchange, users have access to sliders that allow them to fine-tune the model’s output.
Midjourney
This is a business tool. When you sign up, you will get 50 free credits, with an extra 15 free credits granted monthly. Additional credits may be bought for US $15 for 115 credits.
After a contender used it to win a digital art competition at the Colorado State Fair without revealing the image’s method of production, Midjourney developed a reputation for excellence and sparked controversy. The program excels at creating bright, ethereal, and surreal pictures, and its user base has embraced this aesthetic.
The feature is only available through Discord, a popular instant chat network. Prompts are typed into chat immediately. Because chat is public, everyone in a channel may see the prompt you’ve typed as well as the results. It’s certain to perplex readers who are unfamiliar with how Discord operates, which is likely regarded a feature rather than a flaw.
Midjourney is a commercial product that is monetized in the same way as other commercial AI art-generation technologies. Everyone begins with roughly 25 credits, but more need a monthly membership fee. Payment is processed using a web app, which may also be used to examine the photos created in response to your requests.
Craiyon
Craiyon, formerly known as DALL-E Mini, has no direct relation to OpenAI’s approach, and its inventors make the program available for free. The results may take up to two minutes to create and have a poor quality, but nine results are shown at once.
Craiyon is unique in that it uses unprocessed data and makes no attempt to polish, train, or rectify the outcomes. When compared to other tools, its results are frequently lacking, and it struggles with fine details. Human faces, for example, seem unsettling.
The tool has a unique feature. Serving raw results reveals the overall strengths and drawbacks of AI picture production, as well as the difficulties of producing useable results. Because Craiyon does not filter suggestions, it also reveals ethical concerns. Entering an offensive prompt shows how frightening AI image generation can be when utilized maliciously.
VQGAN+CLIP
The increasing success of AI image generators has spawned hundreds of programs that combine complex AI models with a basic UI. One such tool is VQGAN+CLIP, which runs fully in a Google Colaboratory notebook.
It’s worth mentioning since it’s (relatively) simple to use while still providing a look behind the hood. You will be able to see the tool repeat new versions in real time. And, although being accessible via a Colaboratory notebook, the model runs on your own system. Each prompt starts as a blob but gradually transforms into a passable picture.
At least sometimes. The tool’s output is often subpar. It’s sluggish, only shows o