Pros/Cons of AI for the Visual Arts
Discuss the future pros and cons of AI image generation for visual art professionals. Who will lose their jobs? Who is safe? What limitations does AI currently have for artists? Will creativity suffer, and how will copyright concerns be managed in a world where it will be impossible to tell who made what?
If you have something to add to this living article, please send us a note with what you believe to be the issues, benefits, and pitfalls regarding AI-generated artwork. We will add the best responses to the article. No payments are offered, but we will credit your work and provide a link back to your website or social media page. Note: This is not an overview of AI imaging functionality nor a Best/Worst list. There are daily changes that would be impossible to update. Instead, we recommend visiting YouTube where there are hundreds of new videos each month that cover the tools much better than we can.
*****
Added January 14, 2024
In the recent case brought by artists Sarah Andersen, Kelly McKernan, and Karla Ortiz against Stable Diffusion and Midjourney, Judge Orrick decided initially in favor of the AI image generators, writing in his decision that the artists had not proven that a) their artworks were specifically used in the training of the AI models, and b) that the AI generators were creating reproductions of the artists' works that were significantly similar to the point of causing damage to their art careers. The case remains open with the artists given an opportunity to provide more evidence, a process they are currently completing.
Recent revelations in the case include a series of internal emails at MidJourney that discuss the company's ongoing attempts to "copy" the styles of thousands of art genres and artists, thus making it easier for AI users to get likenesses of these artists' styles without actually copying specific copyright-protected artworks.
This along with negotiations between AI content generators and traditional content creators to license data for training AI models is taking place as we speak. Having tested some of the AI image generators over the recent year-end holiday season, namely MidJourney, Dall-E and Photoshop's Firefly, I'll add my two cents.
1. I believe it's going to be impossible to claim copyright infringement on most styles and genres of art. Take cubism as an example. The estates of Picasso and Braque could not lay claim on any "cubist" painting, especially after publicly explaining the visual elements of the genre. Nor could a photographer like Henri Cartier-Bresson claim copyright for a specific style of street photography. While the images generated do share certain characteristics of specific artists, they fall short of replicating actual works. This last point seems to be of considerable importance to the judge, and he appears to be leaning toward the concept of copying an artist's works like a forger rather than an AI artist profiting by riding the coattails of successful artists by copying their styles closely, thus potentially causing financial harm to the original artists.
Breaking it down even further, Picasso could never claim copyright on the process of depicting three dimensions on a two-dimensional plane, nor could he copyright a subdued palette or abstracted portraits. These are among the characteristics of cubism, and they are typical of the works. But they are also employed in myriad ways in other genres that have no other similarities with cubism.
There's an adage in the art world that nothing is really new. Artists have been borrowing from each other for centuries. That includes the joint development of styles and genres. Budding young artists are often taught to make a painting in the style of Artist A or take a series of photographs in the style of Photographer B. The entire concepts of ateliers and apprenticeships were predicated on learning from a master and peers. These factors may also play against the interests of artists in future AI cases.
2. It is clear that the AI generators scraped images of artwork from artist websites and other media to create styles and genres, and that even though these works were displayed publicly on the Internet, this does not suggest that the artists were granting anybody common licenses to use these images in their artwork. This is clearly where the conflict lies, but it appears the cat is already out of the bag on this one. Even if courts decided that the creators should be compensated, how would this be done? How would the courts decide which came first, and what influence percentage could be attributed to it? Proving this is a monumental task far greater than even the droit de suite rulings.
One feature of AI generators that will most definitely need to be modified is the ability to gather the web addresses of images and place them inside a written prompt to create modified likenesses of somebody else's work. In my view, this is direct copyright infringement if permission was not granted to scrape the images. The workaround for AI artists is to create personal databases of images of their own work and use only these to create their AI art. I see a ton of potential for this type of collage or blended work. But let's be clear, this is NOT what the AI image generators are doing, nor are millions of AI users who never attempted to make a single piece of art before cranking up Dall-E to churn out a series similar to their favorite graphic or cartoon artist.
More to come as this case unfolds.
*****
Added January 2024
We encountered this comment after an article on PetaPixel about the spread of AI image generators and their usage, particularly in photographic contests.
The username was @throwawall. We've edited the comment for brevity:
“...the better approach [would be to use] reinforcement learning to validate proximity to a stated goal output one mark at a time, like an artist! Instead [they work backward] from finished works and then fudge the difference with filters (See: Instagram) It’s a design philosophy failure."
Bingo! AI can be and most likely will be a set of indispensable tools for artists and photographers very shortly provided they can produce results that match the goals and intents of the artist as a work is built from scratch one step to the next. Conversely, current AI is simply making guesses to approximate a user’s request. It uses an averaging process where details are lost and quality is compromised for the expediency of providing possibilities. Some admittedly are astounding, but repeatability is still lacking and thus cannot be considered a mastered skill. - JP Paul
I have yet to see a new art style coming from AI art. I see new compositions and content in similar styles, and I’ve seen aggregates of disparate styles. But being trained on the most widely published and most common types of artwork means by nature that there will be nothing truly original generated by AI for the foreseeable future. That’s not necessarily a knock on AI as it also describes most of the art world, but it's worth noting that there are still far more professional digital artists searching for alternative materials and presentation methods to add originality to their work than digital artists who are incorporating AI into their workflow. - D.Pent
*****
Original: June-August 2023
AI image generation has the potential to significantly impact the field of visual art, presenting both opportunities and challenges for professionals in the industry. Here are some future pros and cons of AI image generation for visual art professionals:
Pros:
1. Enhanced creativity: AI can serve as a powerful tool for artists, providing them with new ways to explore and expand creative boundaries. Artists can leverage AI algorithms to generate unique and unconventional imagery, sparking inspiration and pushing artistic boundaries.
2. Overcoming Blocks: Consider AI a useful tool to overcome brain freeze, that blockage many feel when confronted with an “empty canvas or page.”
3. Increased productivity: AI image generation can automate certain tasks involved in the art creation process, such as generating preliminary sketches or assisting with repetitive tasks. This can free up artists' time and allow them to focus on more complex and creative aspects of their work.
4. Accessible tools: AI-based image generation platforms and tools can democratize the creation of visual art, making it more accessible to a wider audience. Artists who may not have traditional training or access to expensive materials can harness AI tools to create and express themselves.
5. Collaboration possibilities: AI can facilitate collaboration between human artists and intelligent algorithms. Artists can work in tandem with AI systems, exploring the symbiotic relationship between human creativity and machine intelligence.
Cons:
1. Job displacement: AI image generation has the potential to automate certain aspects of art creation, which will undoubtedly lead to job displacement for some professionals. Roles that involve repetitive or routine tasks, such as photo editing or basic graphic design are already at high risk. Among the first waves of job market contraction will be for photo retouchers, web image content creators, commercial graphic designers, advertising agencies, stock photo freelancers, visual idea generators, etc.
2. Loss of uniqueness leading to devaluation: As AI algorithms become more sophisticated, there is concern that the proliferation of AI-generated art could lead to a decline in the perceived value of human-created art. The sheer volume of AI-generated content will make it difficult for artists to stand out and differentiate their work.
3. Limitations of AI creativity: While AI can generate impressive visuals, it currently lacks the depth of human creativity and the ability to convey complex emotions and narratives. AI algorithms often struggle with generating original and meaningful ideas, which can limit their application in more nuanced forms of art.
4. Copyright and authenticity concerns: As AI-generated content becomes prevalent, issues related to copyright and authenticity may arise. It is already challenging to determine the origin and authorship of a piece of art, potentially leading to copyright infringement and disputes. Managing copyright concerns in an AI-driven world will require legal frameworks and technological solutions to ensure proper attribution and protect artists' rights. This is particularly important after a couple of recent court cases (e.g., Warhol Foundation) where the concept of “transformative use” of someone else’s images and artwork in subsequent work of another artist was challenged. The holders of the original copyrights won their cases even against larger and more established institutions like the Warhol Foundation.
5. Technical limitations of A.I. image generators: It's sheer folly to form opinions or practices based on the current state of AI functionality. Game-changing advancements are being introduced every week. That said, it is generally agreed as of June 2023 that the images generated by AI are far too small for most visual artists working on physical art which frequently dwarfs the size required for Internet use through computer monitors or smartphones. To date, there is no image creator capable of churning out eight-foot-tall drawings with the resolutions required for adequate gallery display.
The vast majority of AI image generation is being utilized for Internet purposes and game creation, where file resolutions for Instagram pics, website banners, block advertisements, 3d models, and article images use images that are less than 2-5% of the size of large-format photographs and/or computer-generated art destined for physical printing. Granted, there are some excellent image-upsizing utilities, such as Topaz Gigapixel or ON1, that do remarkably well when upscaling high-quality images. However, none of them are capable of taking a 1024x1024 pixel image up to the resolutions required for gallery wall-sized artworks. Artists working with large format prints or mixed media art can breathe easier for a while longer.
Another limitation of current A.I. image technology is the inability to create realistic images from scratch to add to existing images. Programs such as Dall-E-3, Midjourney, or the new Generative Fill for Photoshop can analyze an image to some extent before suggesting three or four optional additions based on the user's written scripts, but the results are often way off the mark, not only as usable content but in terms of size, shape, color, style, etc. It is fun to brainstorm ideas, but the inaccuracy can be a time waster for demanding professionals. I suspect major improvements in this area will be forthcoming within months. At that point, we can revisit this conundrum. - Staff
*****
*****
Added: September 2023
It's important to note that while AI image generation has the potential to seriously disrupt the visual art industry, it is unlikely to completely replace human artists at any time in the foreseeable future. Human creativity, intuition, and the ability to imbue art with personal experiences and emotions are unique qualities that AI currently struggles to fully replicate, even if this is due in part to the human's inability to adequately express the type of imagery they're requesting in short script form.
Even though technology has eventually been broadly accepted as an art form over the past two hundred years (i.e., photography, video, computer art, etc.) many art collectors still value the human touch and that tactile feel that technology continues to struggle to produce due to the technical layers between the artist and the art. What is more likely to happen short term is that the more affordable wrung of the artwork and imagery markets will be overwhelmed by artificial art while those seeking handmade objects, one-of-a-kind pieces, art that creates dialogue, and fine craftsmanship will continue to gravitate toward human-made art, just as many traditional collectors continue to concentrate on oil paintings and bronze sculptures rather than NFTs. - JP Paul
*****
Added Novermber 2023
I appreciate most attempts to protect real artists, but there is also a lot of paranoia with cynical conclusions based on poor information.
As in most industries that were cannibalized by technology, the best in every field will always have work. The group that will prosper moving forward are the professional graphic artists who incorporate AI into their workflow to augment their artistic skills and eliminate deficiencies in their production, for example, speedier idea brainstorming, composition testing, palette adjustments, and the elimination of repetitive tasks. Some might counter that these are not possible with the entertainment-centric initial offerings of image generation, or that programs like Photoshop can do the same already. Again, that's shortsighted thinking. Open-minded artists have already learned how to train image generators on a steady diet of their self-designed elements and images to produce entirely original work with the aid of AI's constantly improving engines.
The last thing I want to see is our collective arts and culture being dominated by tech bros, but the initial stages of any new technology usually feature the replacement of the weakest of the litter. In this case, that will be artists who already rely on stock images for their ideas, or painters who compose their works based on a photo they snapped. One could argue that these artists aren't innovative or creative enough to stay afloat in a highly competitive field anyway, before AI or after.
I believe the proper approach for artists is to understand the technology and see how it might improve their practices rather than continue to bark up the wrong tree. By that I mean make attempts to prove that your work is better than the production of a machine. - Anonymous
*****
For more information, you may find the following articles relevant:
"The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation" - This is a research paper authored by several experts in the field, including researchers from OpenAI. It discusses the potential risks and threats posed by AI, including AI-generated images. You can find it here: https://arxiv.org/abs/1802.07228
"The Dark Side of Generative AI: Manipulating Images, Text, and Speech" - This article by VentureBeat discusses the threats associated with generative AI, including the manipulation of images and other media. It highlights both the positive and negative implications of these technologies. Here is the link: https://venturebeat.com/2019/04/06/the-dark-side-of-generative-ai-manipulating-images-text-and-speech/
“6 artists who where using Artificial Intelligence Before ChatGPT": https://www.artsy.net/article/artsy-editorial-6-artists-artificial-intelligence-chatgpt
“Introducing Composer, The Latest Breakthrough In AI Image Generation": https://medium.com/design-bootcamp/introducing-composer-the-latest-breakthrough-in-ai-image-generation-9a2350e2b9a0
"Supreme Court Rules Against Warhol Foundation": https://www.theartnewspaper.com/2023/05/18/supreme-court-rules-against-andy-warhol-foundation-copyright-case
Federal Judge Sides with AI Companies regarding copyright infringement in a case brought by three visual artists https://news.artnet.com/art-world/federal-judge-sides-with-ai-companies-in-artists-copyright-dispute-2387654"