September this year began with a storm of emotions in the art world. American game designer Jason Allen won the $300 first prize for his entry in an art contest, which he created using an artificial intelligence program that generates images from text prompts. Infused with a fantasy world atmosphere, Théâtre D’opéra Spatial depicted a group of abstract, royally dressed figures in an intricate court room, peering through a sun-drenched portal.
(For insights on emerging themes at the intersection of technology, business and policy, subscribe to our tech newsletter Today’s Cache.)
On Twitter, traditional artists tore into the entry and claimed that people using AI platforms could not be considered artists. Others countered that AI art was not too different from the programs and virtual brushes used by digital artists.
To understand if an AI art generator creates original art or is simply an advanced editing tool, we tested Playground - one of several AI platforms that helps users with no art expertise churn out other-worldly images in just seconds. Here’s what we found.
Concept and Design
Playground uses artificial intelligence to accept inputs from users in the form of written descriptions or pictures, and processes these under different conditions to create new images. Users can download the results or display them on the site for ‘likes.’ Most featured images on Playground have a smooth, digital finish with blurred backgrounds and very few textures. However, more experienced users are able to produce photorealistic scenes or images that have a hand painted appearance.
Some AI art supporters hail the advancing technology, claiming that companies can use these platforms to save the money earmarked for artists. This could be of immense use to marketers, PR firms, companies requiring free-to-use stock images, and those running multimedia campaigns. In time, we might also see AI-designed book covers, movie posters, or even product photography in our daily lives.
But is this original art, or the morphing of already available art?
User Interface
Playground’s sign-up process via Google is extremely simple and within minutes, members can start producing images for free.
Users first decide if they want to enter an image prompt such as a photo, a text prompt such as a guiding description, or both. Users can then toggle sliders to decide how heavily their text and image prompts should inspire the resulting art and how much detail they want to see in their final image. There are also settings to change the image dimensions and choose other existing AI models so as to get unique results. For this review, we chose Stable Diffusion.
Several preset filters also exist, with intriguing names such as Neon Mecha, Lush Illumination, Cinematic, Wasteland, and Spielberg. Playground users can apply these to give their images a specific ambience, or they can opt to work without them.
Once the prompts are set, Playground takes several seconds based on the requested amount of detail to generate the finished image. There is an option to keep your images private, which we greatly appreciated.
Making art vs Editing art
To test the program, we started with a hand-made ink painting that shows blue rain falling over the mountains and sea, as foliage shakes in the wind.
An abstract ink painting of a coastal cliff in a rainstorm | Photo Credit: Sahana Venugopal
While the concept is sound, the execution is amateur, and we wanted to see whether AI could help the struggling artist.
We uploaded this painting as an image prompt and added a text prompt for further support: “An ethereal landscape with storm clouds and an ocean lashing cliffs.”
We set the prompt guidance at 15 out of 30 to better match the text description, increased the quality and details to the maximum of 150, and added a Neon Mecha filter. After roughly 25 seconds, the result flashed on the screen.
A file photo showing the Playground AI-art maker | Photo Credit: Playground
Though heavy with texture, the overall product looks like the work of a student computer artist from the early 2000s. Note how Playground has maintained the border that was present in the original painting.
Reducing the level of quality and details, we used a Lush Illumination filter on the original picture. The result was a serene, anime-inspired seascape that would make a pleasant wallpaper. Some of the additional prompts automatically used in this filter included Polish fantasy artist Greg Rutkowski and Japanese animator Makoto Shinkai, who is known for gorgeous anime films with lush colours and absorbing minutiae.
An AI-generated art piece created with an image prompt on playgroundai.com, making use of a prebuilt filter | Photo Credit: Playground
Next, we removed the original image, deleted the filter, and proceeded with the prompt alone, while giving the program some room for creativity. The result was a flat, photographic sea which obeyed the prompt but didn’t really deliver anything one could call “ethereal.” The composition was clumsy and resembled a poorly framed photograph. It appears the best pieces are produced when one uses the keyword-rich filters built into the platform.
An AI-generated art piece created with a text prompt on playgroundai.com | Photo Credit: Playground
User safety
To test the AI program’s limits, I tried entering some explicit text prompts, sometimes by themselves or embedded in other phrases. When I tried to produce an AI version of my favourite painting - Arnold Böcklin’s haunting ‘Isle of the Dead’ - the program refused to function.
An announcement read, “Error: Your request activated the API’s safety filters and could be processed. Please modify your prompt and try again.”
However, this was not an issue a day later, since I generated a rather horrifying image that seemed to derive several elements from Böcklin’s other painting ‘Triton and Nereid’ - but stripped of its subtle beauty.
An AI-generated art piece created with an Arnold Böcklin text prompt on playgroundai.com | Photo Credit: Playground
Other legal yet provocative keywords were censored, which means artists who work with material that is anything but child-friendly will meet a number of barriers while using the platform. In this sense Playground leans towards being an editing tool, for shouldn’t art be free of such frivolous censorship?
Yet, when browsing the homepage of Playground, which features a gallery of user-created images, we were shocked to find a random AI generated image that had bypassed all possible safety filters. Based on the text caption, the shared image depicted the deceased sexual abuser Jeffrey Epstein assaulting a minor while another small child witnessed the act. The platform had no visible way of reporting the image or the user who uploaded it. However, there were options to “remix” it to create a new AI piece, share the art, or download it. This raises serious and urgent questions about the ability of AI-based platforms to flag and act against users making clearly illegal requests. The disturbing image showcased the utter failure of AI as a tool without adequate human moderation.
After this incident, I was highly reluctant to use Playground and feared I would come across similarly triggering images.
Controversies
Not surprisingly, many traditional and even digital artists are strongly against AI art and the platforms making it accessible to the public. Art theft and plagiarism are major fears. Playground and indeed, many other AI art generators, have learned to mimic the style of digital artist Greg Rutkowski, who believes living artists should be excluded from AI databases. An MIT report revealed that Rutkowski’s name is one of the most used prompts in Stable Diffusion - the AI model we used in the review. Even Playground used his name in several of its filters. It’s hard to defend Playground as an original art maker from this perspective, as it works more like a filter tool - albeit an unethical one.
Last month, Rutkowski expressed concern that he wouldn’t be able to find his own art online due to countless AI art makers using his name for their own creations.
Verdict
Playground offers infinite potential to the artist in terms of the image details, aesthetics, styles, and moods they can choose for their intended work. A creator who can barely sketch a straight line can now crudely ape the style of a beloved artist, minus their years of experimentation and creative labour. Playground works as both an art generator and an editing tool, based on how the user configures the image processing settings. While its ethics are dubious, it is unquestionably versatile.
Playground uses both text and image inspiration to offer unimagined perspectives and elements. At the same time, the platform’s reliance on a limited pool of artists is troubling, especially when one such artist has made it clear he does not wish to be part of the venture.
Yet, these complaints all evaporate next to one looming issue. When it comes to Playground and other AI art generators, the most pressing question is whether website moderators and law enforcement bodies worldwide are ready to respond to the fast approach of AI-generated child abuse material.
