One of my works in progress is a science fiction novella about the creation of an artificial general intelligence (AGI). The story features a scene where the human programmers are amazed that the AGI can create original artwork of any kind on demand. I wrote the first draft in 2018. Yes, just four years ago an AI that could create art seemed like a speculative bit of futurism. Now it appears I will need to revise that scene, as what was “sci-fi” then is now just “sci.” Reality is progressing faster than I can publish science fiction.
When I wrote this post about DALL-E last May, I had only seen others’ generative-AI creations; I hadn’t gotten the chance to create my own AI art yet. Now I have and am utterly addicted. There was much hype around AI image generators like DALL-E and Midjourney when they were first released. Usually when something is hyped that much in the media it is overblown; the reality is far less dramatic. But after DALL-E became public, plus the release of the free and open-source Stable Diffusion, I have had the chance to create my own AI art (thousands of images at this point—and counting). While the initial hype was quite high, I would venture to say it was not nearly high enough. Most people still don’t realize how significant generative AI is/will be. In the future, people will look back at the world pre-AI art as a distinct, unrecognizable time. Generative AI is a total game-changer, an artistic singularity.
The world will be forever changed with AI image generators (text generators too, but more on that in a future post), in ways both good and bad. From now on, AI art will be used everywhere: book covers, movie posters, blog posts, music albums, concept art, commercials, corporate logos, YouTube videos, comic books, social media, memes, and more. Quite soon, people will see more AI-generated art than human-generated art online (and off). There is an obvious downside to that, but there is also a massive upside. [Naturally, AI generators will also be used for propaganda, pornography, and spam, but in this piece I want to focus on the potential positive use cases: art.]
The Art of the Prompt
Using AI image generators has made me realize that at least half of creating art is choosing what to create. That is what you are doing when you type a prompt into DALL-E. Despite their enormous artistic prowess, AI generators will not produce a creatively interesting image unless the human types a creatively interesting prompt (aside from random luck). I’ve always been a visual thinker with a rich imagination. When I was younger, I would draw, sketch, and doodle often, but I eventually stopped and rarely draw anything anymore. I since redirected my visual creativity toward photography and digital image editing with programs like Photoshop (or its free open source alternative, GIMP).
Whenever I try to draw, it never comes out as good as I imagined in my mind. But using AI, I can describe an image from my imagination, and the AI can paint it for me—with much more detail and skill than I could ever muster. Of course I could simply devote more time to developing my drawing/painting skills, but that time would come at the expense of time I could be writing—a skill that comes more naturally to me. Writing is my true artistic talent, but I’m happy to use AI to augment my creative skills and expand my art into other, more visual mediums.
Being able to instantly bring a fantastical image from my imagination to fruition feels like magic. Sometimes I’m shocked at how good the first AI images come out based on my initial prompt. It feels like my imagination come to life. But more often, the images don’t come out quite how I imagined. They are usually different—sometimes in odd and interesting ways—though more often in odd uninteresting ways. So then I need to go back to the drawing board—or the writing board—and craft a better prompt. Though with certain prompts, especially involving people and animals, they always come out distorted. AI seems to have an especially difficult time generating skateboarders for some reason. I’m sure AI will get better at generating people with time. Though for now, those eerie distortions of humans can actually be helpful if you are intentionally trying to create horrifying art. AI generators could produce the next H.R. Giger, creating the most hauntingly disturbing images of cosmic horror.
I’ve already started using AI to generate images for this blog and Time Zone Weird. It is great for “Future Fake News” reports. But another thing I’ve been using AI image generators for is creating concept art for the long-form fiction stories I’m currently working on. Images of characters and scenes from your story can be inspirational and help set the mood when writing. Some AI concept art has inspired me to change the story to match an element randomly produced by AI, and other AI images have inspired me with completely new story ideas. I just need to be careful not to become too obsessed and spend more time creating concept art than writing the actual story. Stable Diffusion can lead to unstable procrastination.
I previously speculated that “prompt writer” could be a new job in the future, and I am now even more certain that prompt engineering will become an extremely valuable skill. Crafting a prompt is an art in itself. You need to be able to write clearly and be able to describe things well—but also have a vast knowledge of art and art history, able to cite the artists, movements, and styles you want to emulate. While human artists are at risk of being replaced by AI generators, it is those same human artists who are best situated to become expert prompt writers—because they will have the most comprehensive knowledge of art. They can cite references and descriptions in their prompts that most other people will never have heard of. Plus they can use their own artistic skills to edit and improve upon whatever image the AI generator spits out. As of now, most AI images still have some flaws that need the touch of a human hand to polish into a finished product.
While I appreciate art and have a keen eye for aesthetics, I never studied art history deeply, nor am I familiar with the contemporary art scene, so I don’t have a comprehensive vocabulary to describe art. I just know what I like when I see it. While I can use AI to keep creating images until I see something I like, it would be more efficient and effective if I did have knowledge of art, both historical and contemporary. Those who study art can be more precise in their prompts, including details of artists and styles, then they will get better results from the AI. Those with a comprehensive vocabulary of art can become prompt-writing experts. Art History degrees may suddenly become much more valuable. Every major company will be looking to hire AI whisperers who know how to type the best prompts.
Others, like myself, who can’t afford to outsource prompters must learn more about art and artists, both past and present. I should have studied Dali long ago, but DALL-E will spur me to do so now. Contemporary human artists will benefit from the proliferation of AI art as well. People using AI generators will need to learn the names of current artists they like in order to mimic their styles, which will ultimately lead to more exposure and opportunities for those original human artists. When mediocre art is ubiquitous it only makes exceptional art more valuable.
Garage Kubricks and AI Movies
As impressive as AI art generators are now, we are still in their infancy. They will only get better from here on out. Imagine the future possibilities of DALL-E. There are already text-to-video AI generators, and it is only a matter of time until they can create feature-length movies. I predicted such a technology in this post from 2020 and even earlier in this post from 2019, but now that speculative future seems much closer and more inevitable than ever. Soon you will be able to write a screenplay (or use a GPT to write it for you) then feed the script to an AI that can produce a professional-looking movie. You can select which actors (from all of film history) you want in the roles, including every possible hybrid combination between them. You can choose the composer(s) you want to create your musical score, choose a cinematographer, a director, and any style of visual effects. On my Substack, Time Zone Weird, I speculated in a “Future Fake News” post that such an AI movie generator might be called “Q-BRiC” or “Quantum Bits Rendered in Cinematography.” Whatever they are ultimately called, AI text-to-video generators will be an even greater game-changer in the future of art and culture.
One of the downsides of movies is they are so expensive to make that it creates a natural gatekeeper. Very few people/companies can afford to finance a movie, so those that do have great control over the types of movies that are made. AI video generators will democratize filmmaking so that anyone with a computer and internet connection could make a movie. This will result in a deluge of crap, as has happened with self-publishing books. But, as with self-publishing, it will also result in artistic geniuses who otherwise never would have gotten the chance to bring their passion project to fruition and easily share it with the entire world. Success in Hollywood filmmaking today is based on who you know, but success in AI-generated movies will be based on what you know—or how creative you are. The only limit will be the prompt director’s imagination.
What if Stanley Kubrick never had to worry about budget constraints or actors not performing exactly how he wished? Future “Garage Kubricks” (a term coined in 1999 by science fiction writer William Gibson for auteur film geeks able to create movies entirely in their garage with digital technology) will be able to create their cinematic masterpieces with zero studio interference, no notes from producers or focus groups, and no need to cut costs for special effects. With budget no longer a constraint, anything you can imagine AI can animate into film. We will see a renaissance of filmmaking, a new golden age of movies better than the 1970s, or any other time in history. Auteur directors will have the ability to micromanage every aspect of the filmmaking process and use AI to create each frame precisely as they desire. Movies are unique among all mediums of art because they combine all other mediums of art into one (writing, photography, dance/choreography, drawing/painting, music/singing, costumes/fashion, etc.). AI movies will become the ultimate art form, their human prompt directors the ultimate artists.
Before you worry about human actors losing their jobs to deepfakes, actors can benefit from AI movie generators as well. Actors today who struggle to get cast in leading roles could star in their own AI-generated movies. They can film themselves acting in front of greenscreens in their bedrooms then have AI superimpose them into scenes with other actors (human or AI). Prompt directors may even hire human actors for certain roles, preferring them to AI-generated characters. The same goes for composers, visual effects artists, costume designers, and every other name you see in the credits at the end of a movie. The top artists in each craft involved in filmmaking will likely be better than the AI generators who copy them, so prompt directors may want to hire those top human artists to contribute to their films. The best AI movies may be a collaborative effort, but it will also be possible for individuals to make perfectly competent AI movies all by themselves.
There will likely be an uncanny valley where astute viewers can tell if a movie is AI-generated, just as today many viewers can spot CGI and prefer practical effects. The best use of AI movies might be animation where you are not trying to attain perfect verisimilitude. Animated movies today are quite expensive, but with AI the cost could drop to zero. Then anyone could make an anime or Pixar movie. Regardless of how realistic AI movies become, there will remain a subset of future viewers who prefer old-school “craft movies,” made entirely by human effort with no AI.
If you think there’s too much TV/movie content now, just wait until AI can produce video. There will be millions of new movies created every day. You could even create fan edits of existing movies with new actors in the roles, different directing styles, or an alternate ending. Use AI to remake The Matrix starring Will Smith. Or remake The Wizard of Oz in a modern Pixar-style computer animation. Or put Batman in The Godfather—because…why not? Somebody will inevitably generate Jodorowsky’s Dune. You could feed the AI a novel and get it translated into film form. H.P. Lovecraft’s At the Mountains of Madness can finally be put on screen—millions of times in infinite variations. I can create my alternate version of Star Wars Episode 9—and Halloween 78. And I can adapt all of my books and short stories into movies—with various iterations directed in different styles, from Christopher Nolan, to Tim Burton, to Stanley Kubrick, to David Lynch. An AImdb.com will arise to rank the best AI-generated movies. Film critics will become ever more valuable to help separate the wheat from the chaff. Choosing among the seemingly endless streams of tv/movie content available today is difficult enough, so curators will be even more essential in the future when the content will be literally endless.
Movies and television are becoming oversaturated today due to the preponderance of internet streaming services. What is happening now in movies is what happened to novels twenty years ago with the advent of ebooks, print-on-demand, and self-publishing. There became too many books available for a single novel to become watercooler material that everyone read and talked about. Less people read now and those that do delve into various insulated sub-genres. Movies and tv are in a similar place today, as there is too much competent content, and people don’t have the time or attention spans to consume it all. There are no longer big hit movies/tv shows that everybody sees. Instead, tv/movies have become much more niche. Hence it’s rare that you and your co-workers at the watercooler have watched the same tv shows—though you can easily find a fellow fan from the other side of the world via the internet who shares all your obscure interests. In the future you can chat with them at the metaverse watercooler.
AI-generated video will accelerate the fracturing of tv/movie fandom and popular culture in general. The future of movies might look like YouTube today, where there will be millions of creators releasing their own AI movies in increasingly specific sub-genres. Plus YouTubers will use AI to create short-form videos in infinitely various ways. So many YouTube videos today use the same stock footage because most creators don’t have the time or money to film original content. Unique videos created by AI—even if not as good as human footage—will be better than seeing the same stock videos over and over again.
Creators have already begun to use AI art on YouTube today. But human artists are not being replaced because of this. These are videos that otherwise would not have been made (or would have been made with bland stock footage or just the talking head of the creator). These are YouTubers who can’t afford to hire camera crews or human animators to create original footage anyway, and the stock video they use is not exactly high art. The YouTuber is the artist in the medium of YouTube, so creators will use AI generators to improve their videos—but there will remain a human director at the helm. AI art will supercharge the creativity of human creators.
No More Gatekeepers
Initial barriers to AI movies will be time and storage as video takes up more hard drive space than text and images. Both rendering, editing, and creating an AI movie will likely take more time and effort than an image (including many more text prompts), so less people will create AI movies less often. You could potentially write a short prompt (like Star Wars meets Goodfellas directed by Quentin Tarantino), but the result might not be a compelling 2-hour movie that makes any narrative sense. An AI generator could probably create a short scene from such a mash-up movie or a trailer (as some YouTubers currently do). But the total prompts needed to create a great feature-length movie may require more words than an average screenplay today. Serious AI movie directors will spend countless hours fine-tuning each frame of every scene in their eventual film.
Some movie directors meticulously frame every shot of their movies to look exquisite, while other directors just point the camera at the actors. Likewise, certain movie watchers appreciate the directors who focus on cinematography while other viewers don’t even notice such details. Artistic auteur directors will spend more time intricately creating AI movies and have a heavier hand in the final product, while other hobbyists may simply type a few prompts and let the AI do all the work. A movie made completely by AI may be good enough for layman viewers, but film geeks will veer toward the AI films finely crafted by auteur human directors (the Garage Kubricks), as film geeks like myself do today.
Despite some initial bandwidth limitations, once the technology is available, there will still be a lot of AI movies created. Consider that in 2022 there is almost one million hours of video uploaded to YouTube each day. Just as most YouTube videos are not worth watching, most AI movies won’t be worth watching either. But even if only 0.1% of AI movies are good, and a million new AI movies are made in a day, that would leave you with 1,000 4-star AI movies to catch up on—daily. And the number of AI movies made each day could end up being more than a million—and the percentage of quality AI movies could be well over 0.1%. There will be a vast desert of rotten tomatoes, but lurking amid them will be cinematic gems—too many for any one person to ever watch. But that is not a bad thing. It just means each individual can go deeper into their personal interests and see exactly the movies they want to see. The future of Netflix might be recommendation algorithms automatically using AI generators to produce new movies based on the previous movies you liked. You might be the only person in the world to ever see the particular movies in your queue.
AI movies will completely destroy the Hollywood movie industry—and that will be a very good thing. With budget no longer a concern, the gatekeepers will collapse. When anyone with a laptop can create a Hollywood-caliber movie by themselves, what is the purpose of Hollywood? AI movies will erase the incentive for studios to make lowest-common-denominator movies to try to appeal to the largest audience (such as the MCU). Post-AI movies there will no longer be mega-blockbusters and remakes of reboots. It will be impossible to control IP when anybody can create fan films with AI movie generators, even if those movies are non-monetized (which they probably won’t be). People can use AI to make both woke and non-woke versions of every possible movie—and audiences can choose which they prefer. Fans will edit and iterate upon each others’ AI movies as people do now with internet memes.
There will be thousands of cuts (at least) of every popular film. Take, for instance, an adaptation of a book like Blood Meridian by Cormac McCarthy. Some prompters will adapt it directly: a 12-hour miniseries that includes every scene and every line of dialogue precisely as it was written in the novel. Fans of the book might want to see this loyal version while other AI movie directors will streamline the story for more action and drama, cutting out the “boring” (to them) parts of the book. And every director will cast their films differently. A viewer might even take an AI movie created by someone else and re-cast one or more roles—or add a different musical score, slightly alter the directing style from Ridley Scott to the Coen brothers, or employ different special effects and lighting, or different editing techniques (quick cuts vs. long takes). Not only can any movie be made with AI, but any version of any movie.
Remakes, sequels, and re-imaginings of existing IP will likely be the first and most common use of AI video generators. These types of movies will be easier to make because of the pre-existing material the AI will have to work with. What will be tougher for AI to generate is completely original films. Hence to have any chance of competing with AI movie generators, Hollywood—or any human filmmakers—will have to make original movies. The current system incentivizes making redundant movies exactly like the most successful movies from the recent past, but the future post-AI text-to-video system will incentivize making movies so unlike existing IP that AI will not be able to generate it with current data sets. Just imagine a movie business that values imagination.
Due to the large supply of AI movies and the low demand, most AI movies won’t make any money at all. When movies cease to be big business—a way to make hundreds of millions of dollars—those who pursue filmmaking will do so only out of love for the art. Even with the aid of AI, it will still take skill and effort to create a great movie. There are potential future auteurs alive today who will never convince Hollywood to finance their dream films. There are even current auteurs who struggle to get their passion projects green-lit, and instead work on mediocre movies just for the paycheck. What AI will ultimately do is give human cinematic geniuses the complete freedom to create their personal masterpieces. This is what every new technology has always done: empower human artists to make more innovative art.
I once imagined that if there was a heaven, Stanley Kubrick would be there with endless time and an unlimited budget to create whatever movies he wants. That daydream will soon be a reality on earth for future Garage Kubricks. As a film buff, I cannot wait. I just need to finish editing that story of mine before AGI is actually created and the entire premise becomes obsolete.
All images in this post (except for the Spiderman meme) were created with AI. Here are some links for the AI art generators I use:
- Stable Diffusion
- Bing Create (my most recent favorite)
P.S. I will also be posting some of my AI-generated art on the Time Zone Weird Instagram account: @tzweird