Denoise: How Scary are AI Art Generators?

Neural network-based text-to-image generators have been around for quite a while now, but only in the last year or so have they gotten good enough to capture the general public's attention and force us to ask real questions about the value of art and artists.

One big reason they have improved since the days of blurry 240p results is because of the vast amount of data, i.e. images, that they’re trained on.

Many of these images are copyrighted by companies or individual artists and can be closely replicated by the generator if given the right prompt.

The question of whether or not the practice of training an AI on copyrighted data is legal is still up for debate in court, but regardless of the technical legality, many artists are understandably furious that their past work is being used to render their future services less necessary.

We at CG Cookie have generally not shared our thoughts about AI art for two reasons.

Firstly, we generally make Blender tutorials and there was nothing about the topic we wanted to make a tutorial about.

Secondly, our team is made up of a variety of humans with a range of experiences and biases who don’t all share the exact same view of the topic. Recently, however, Kent, Tim, Kyle, and Dr. Denning got together to express their thoughts while recording an episode of our Denoise podcast.

Continue reading for a quick summary, or listen to the whole conversation below. 👇

This post is from episode 5 of the Denoise podcast. Listen on iTunes, Spotify, Google, Amazon, and wherever else you find your podcasts.

What is AI Art?

“AI” is a fairly vague buzzword, but when people talk about AI art, they’re specifically referring to a neural network that can create entirely new images based on the style and subject matter of the millions of images it was trained on.

The most famous AI image generators are DALL-E, Stable Diffusion, and Midjourney, but there are more being created all the time.

One thing worth noting is that since AI pulls all its data from existing images, it can only create mashups of what currently exists with a bias towards what is popular.

Even if feeding copyrighted images into an AI is determined to be legal, it is still illegal to pass off a sufficiently similar image or likeness as your own creation.

For example, if you tell an AI image generator to create a licensed Star Wars character and you try to sell the image, Disney may not be able to go after the AI company, but they could decide to go after you.

In some cases, you could be breaking copyright without even knowing it. This is one reason why the US has decided that the output of an AI generator cannot be copyrighted. Use with caution.

Are artists really losing jobs because of AI?

Generated art is not going to replace artists any time soon fully, but it doesn’t have to replace all artists to have a substantial impact on the industry.

Even if each company that uses AI is able to hire just slightly fewer artists, the already competitive field could become incredibly difficult to break into.

Many larger companies have banned the use of AI due to copyright concerns, but some have gone a different route and trained in-house AI models based on their own artists' work.

Side jobs are also harder for artists to come by, since commissioned artwork related to rendering family, pets, or D&D characters in a painterly style is what gets asked for the most and is also exactly the kind of thing AI excels at.

According to concept artist Tim von Reuden, the larger companies needing fewer employees and the commissions from the public drying up are together, making it hard for more than a few artists to pay rent.

Do people care if a human made it?

There will always be some people who love the craft and the ‘old fashioned’ way of doing something, such as vinyl collectors who enjoy a good record on last century’s equipment, but most care about the quality, quantity, or convenience of the final result.

According to our very anecdotal evidence of asking family and friends, many folks wouldn’t think differently about an artwork or film if AI created it.

As 3D artists, we work at the cutting edge of combining human creativity with the power of machines. We ‘cheat’ 3D animation daily by interpolating, simulating, and rendering rather than drawing and painting.

How much of my results today can I say I came up with vs. the computer?

It’s already a spectrum. Somewhere along it is a line between meaningful human creativity and soulless machine outputs, but the meaning of a piece is ultimately up to each individual viewer and not necessarily its creator.

Are we excited or worried about the future of AI?

Yes. AI could be a great solution to many of the tedious tasks we currently have to complete manually and could allow individuals to create more complex scenes than was ever possible by a single person before.

At the same time, if that power comes at the expense of our fellow artists, it isn’t easy to be thrilled about the results.