The robot uprising may still be far away in the future, but AI is beginning to look scary for some of us. The advancement in Artificial Intelligence technology has led to some interesting development, on the way to developing a general purpose A.I researchers have come up with a tool known as an image generator.
You may have heard it before, it’s almost impossible to avoid lately. Yes, we’re talking about the rise of tools such as Dall-E, Imagen, Midjourney or even Disco Diffusion.
With differences in their underlying tech and dataset, those A.I tools pretty much does a similar thing. Basically, you type in a sentence, and then they will spit out an image or two (or ten).
The text or sentences are commonly called a prompt. It could range from the mundane (a sleeping cat on the floor) to the absurd (a cat riding a unicorn on the moon) and the A.I will give you an image.
Some of those images can sometimes be quite impressive. An artist even won a competition by submitting an artwork created using such tools. (We’ll get to this in a moment).
The rise of tools like Dall-E and the like has given rise to the usual fear of technology replacing humans. We have seen this fear during the advancement of photography and even during the industrial revolution. But so far, those fear has never really manifested into the doomsday scenario that some were afraid of.
But, you may say, something is different today. The tools, the A.I., isn’t just a new way of doing something, it could very much replace the work of a human being.
Yes, that might be true. But we shouldn’t just go and ban or hate on this upcoming tech.
Where Did The A.I Comes From?
Okay, first let’s back up and have a quick history lesson. Where did all this A.I. actually come from and why were they created.
The A.I. race has been going on for some time now. But we can argue that Google’s AI win over a human player in the game of Go, in 2016, really kicks things up a notch. Previously, headlines about A.I. achievement were usually won by IBM.
The race got hot quickly with Amazon dan Microsoft also looking at A.I. seriously and publicly. It was not a coincidence that a year earlier, Elon Musk, Sam Altman, and other tech investors pledge their money to OpenAI. This formerly non-profit organization is currently backed by Microsoft.
In June 2020, OpenAI announced GPT-3, a language model trained on trillions of words from the Internet. This system’s aim is to understand everyday language, also known as Natural Language Processing. From this, DALL-E was born in early 2021.
DALL-E’s name is a combination of the beloved robot character WALL-E and renowned painter Salvador Dali.
What Does The A.I Do?
DALL-E, Imagen (from Google), and several other similar systems are really buzzing on the scene right now, mid-2022. Each has a similar way of using it, such as the aforementioned “text to images” generation.
Other common features are “inpainting” and “outpainting”. Inpainting is basically the ability to fill in missing parts inside an image. While outpainting could refer to the ability to generate images outside of the current image area.
For example, if we delete the head from an image of the Monalisa painting. We can use “inpainting” to draw something else in its place, according to the prompt given. While if we use outpainting, the ‘canvas’ is expanded and A.I. is used to draw on the new area, according to the prompt.
How does the A.I able to do this? In short, the system learns from an abundance of images on the internet. This dataset consists of an image and its description.
So, every time someone inputs a prompt to the system, it will ‘learn’ from the dataset to understand what is the text about. And then, through a method known as diffusion, it will generate an image.
Now, diffusion itself is pretty cool. If you ever used Midjourney, it will somewhat show how this work. At first, the system will just create noise, then it modifies that noise slowly until it resembles something similar to the text given. It does this several times until it finally creates the final image.
Jason Allen’s painting was created using Midjourney.
Should Artists be Worried?
In short: no. This is just like the invention of any other creative tool. This may render some jobs obsolete, of course, but it may also create new jobs or new opportunities.
So, about that case where an artist won a prize for submitting a generated artwork. Here’s the story.
Jason Allen, as quoted in many places, was announced as a winner of the digital art category at the Colorado State Fair fine arts competition. His submitted work was titled Théâtre D’opéra Spatial and has been described as a sci-fi/fantasy scene of epic proportions showing a portal and some people. Allen used the A.I. Midjourney to create the image, and Gigapixel, also an A.I. based system, to upscale the work before printing it on canvas.
Many artists were outraged at this decision. But the organizer and judges stood by their decision and even welcomed the conversation it created. Allen himself explained his work.
He described the process as trying many prompts and fine-tuning the prompts, before finally reaching that final image. In total, he created 100s of images, and the process of fine-tuning the prompts and curating the generated images took weeks.
Allen’s win actually highlighted some things about the role of A.I in design, arts, and creativity. At its current iteration, all those tools still need humans to drive them. Just like a camera needs a person to operate, A.I. needs a human to create something valuable to another human.
Another thing that should be noted: the A.I. still used datasets from already existing images. Most systems are able to generate art based on previously known styles, such as Picasso, Basquiat, or Rembrandt. But it does underscore the fact that there needs to be a Picasso or Rembrandt in the first place.
Screenshot from Draw with Jazza, comparing an A.I generated art (left) and Fiverr commissioned art (right).
A.I versus Fiverr
One common thread of discussion about these A.I. tools is: will it replace freelance jobs? Especially for casual and everyday tasks, not replacing fine arts and masterpieces.
An example where the A.I. can be used is when a company, or someone, needs a stock photo for a brochure or just to illustrate a blog post. These A.I is said to be more than capable of handling such a task.
The art vlogger Josiah Alan Brooks, otherwise known as Jazza, created a video where he compares art generated by A.I with art commissioned through popular service Fiverr.
Some A.I. arts generated by Jazza were superior to the Fiverr commission. It was even discovered by Jazza that some art from Fiverr was pulled from existing stock photos.
But in one case, an art created by a human on Fiverr was able to illustrate the prompt in a more pleasing and accurate way. Even when that artist didn’t demand so much in price.
Jazza seemed overall unimpressed by the A.I capabilities and was rather disappointed by so much low-quality work on Fiverr. This actually illustrates how you need to be really selective when commissioning work on Fiverr.
Choosing works from Fiverr verified Pro is one way to ensure you have the best work. And also look at the portfolio of works they have created. Don’t forget to communicate clearly what you expect from the work to be done