My original full post with the images, musical links, and the novel itself can be found on Royal Road in the Art Guides section under the name AI-Graft, the art of collage in an age of AI, a glimpse into the future. (Sorry, can’t put the links and more than one image as a new user.
First, a bit of house cleaning is in order. I do not claim that a single prompt is art. I do not think that AI is completely harmless. I’m quite nuanced on these subjects because I’ve been expecting it since I started writing my novel in 2008. I prepared for this age because I knew it wouldn’t be easy, but it would also have some incredible potential applications. It is a technology, and as for any other technology, it has its flaws, dangers and benefits like splitting the atom. It is, however, potent. Using it for good when it is ubiquitous is important to counterbalance the negatives it will inevitably bring. History is already in march, there is no turning back the clock, AI is in the world, now the question is how to use it properly to sustain the existence of human artistic expression.
This is my partial answer; it is not absolute, it is experimental in nature, and it is only the first step toward this new MEDIUM. In my use, AI is not the artist, it is one of the media I employ to transmit my personal, long-running artistic intent. This general guide is not a technical guide, it is a glimpse of what we can do as humans that AI currently can’t and will probably need a few more iterative steps to even make a near similitude.
What is AI-Graft: AI-Graft, as I like to call it, is the art of collage of, in my case, hundreds of coherent AI images and songs converging around my own work, my novel. The novel is NOT AI-generated. AI does assist me in creating, though, but through indirect means, I feed the content of my work every day and ask for discordant and concordant elements so I can prune and complexify it at my leisure. AI is fundamentally a symbolic resonance cavity. It doesn’t produce something, it resonates themes and ideas, but keeps them in the shape of its training. The more complex meaning you infuse in it, the better its qualitative output. You can create a feedback loop in terms of the symbolic quality of your narrative by asking it point-blank what themes and ideas it sees in your text. Then the feedback loop is you either reinforcing these themes or removing them to prune your symbolic if you do not want to support that thesis, and it crept in by accident.
What AI is good at: Single prompt, small element creations. Images and songs, IF properly guided, and an extensive process of iteration and pruning is applied. Do not let your artistic intent and your vision be diluted by easy but imperfect results. Keep at it.
What is AI bad at: Long narrative lines, subjective taste, and keeping track of everything. Basically, its attention skills are horrendous, don’t know what’s good… this is why it gives so much AI slop if the user is lazy. As the user, you need to be firm and structured and not go the easy route of just accepting anything. Keep pushing, understand its limits, but and that’s important, sometimes the AI, like a human collaborator, will add meaningfully by accident or through resonating with your larger themes, so sometimes they do surprise you with creative ideas.
Here is an actual example from my work.
The idea of the overall structure and concept of the image was mine, the necklaces, my MC in the prison belly. It was to show she was caught in her primary need, hunger, but also showing it was making her a beast. This is where AI came in with a boost… my idea was to show her helpless, more like “caught princess” than this position, the beast was in the first sketch much bigger, but the perspective in this attempt here makes us understand that she is positionned as to be also in its legs and not just in the stomach. In essence, the perspective reinforces the idea that Kate is the beast driven by her belly, that the cybernetic monster is Kate and not an outside entity, an extension of her, and she’s trapped. My core idea was respected in an unexpected and ultimately more satisfying way than I initially conceived.
Keep in mind, this is one of my first; the models are getting better, and it will be remastered later once tech moves forward. It had already gone through a few iterations. The main problem with AI music currently is syllable pronunciation artifacting and how hard it is to correct anything once it is gelled into a coherent whole. ).
How the song was made, I chose this one because I didn’t make the full lyrics, unlike some other songs. How do I proceed? I wrote my novel, again by myself (only symbolic feedback loops I mentioned, and I didn’t even use it for that at the time, as AI did not even exist then). I fed it the whole into AI with every contextual piece I could give it, then asked it what subject it was envisioning for a song related to the chapter. I talked about it for a while and decided through brainstorming with the most appropriate to my sensitivities. Then I asked for a bank of appropriate music styles and lyrics. Then I reworked the lyrics fully, adjusted the rhythm, tweaked the meaning, and fed it to another AI with the guidance for style. From there, I tried over and over with different style prompting, retweaked the sections of the lyrics that didn’t crystallize right consistently while keeping the meaning (sometimes removing or adding a word or changing it).
To be clear, this song is one of the numerous songs in my book. To give you a sense of scale, I expect over 150 chapters with a minimum of 1 song per chapter. Each of these songs will have its own cover art for the single with a coherent (though varied to represent multiple cultures and genres) art style. I’m also adding to this 2+ images directly inside the chapters to support the text.
Ultimately, the idea of AI-Graft is to make a collage, like let’s say people who cut gloss magazines and assemble them in mosaics. Each AI parts are grafted together to create that mosaic of meaning that each of them does not have sufficiently individually to be considered art. I’m creating themes and symbols by what I choose to keep. So the album covers I compose, as an example, have a vintage feel, to give them a sort of gravitas of nostalgia, of mourning a world lost.
The trick is to lean into fractal storytelling (concepts that are aesthetically pleasing and infinitely replicable in every and all scales without constraining your creativity too much).
As for the in-text images, I’m adding well, I’m trying to find key moments where an image is useful. From character introductions, to action scenes and through conceptual elements. Here is an example of the ones currently in chapter 1 (more will be added later, this whole process is fresh to me, a few weeks old, and the whole idea only crystalized in my brain yesterday night… or should I say early this morning, so bare with me this is not the final result by any stretch)
All this to give you an idea of little Paris, Paradise City, Moon. The ambiance, the tones, without using too many words. To cut through the clutter. To focus on what ultimately matters, characters, story, philosophy, action, consequences… Show, don’t tell extreme edition.
Writers and other large-scale holistic artists have an advantage in this future sphere. Something AI has a HARD time doing is coherence and taste. So one one-hit single musicians and people who make visual art in single frames will have a much harder time competing; the former have a field advantage by completing our own vision and, for now, an inaccessible skillset to autonomous AIs and just use them for extra storytelling leverage. The undervalued skill of coherent conceptual world-building finally gets to shine.
Think of our role not as the grunt worker of art but as the art director, the realisators… and well, for now, the writers. It is YOUR vision that needs to shine through. You need to sharpen your tastes and your own skills because someone who knows what questions and jobs to ask the AI has a large field advantage in the output. Garbage in, garbage out.
Here are the fields that helped me the most while speaking to AI and building my world conceptually:
Symbolicism
Art history
Philosophy
Mythology
History
Music theory
cultures (Varied and in-depth)
Psychology
Sociology
Visual composition structures in art (golden ratio, stuff of that nature)
Architecture
Knowing multiple languages (to get around the symbolic barriers of a single one… when an AI struggles, ask it the same question in another language and it will give different outputs)
To get the most out of AI, you need to know as much as you can about everything you can to have a sort of map of everything you want to portray, so you can point in the right direction and know if it is right when the result comes out. You can get second opinions and help from it, but it is a crutch and, like walking on crutches, it’s a mixed bag result.
Also, it is crucial that you keep sharpening your voice and artistic intent as a writer and an artist in general, and that you create your own aesthetic pattern. The better you are at it, the better the final output. Yours are the choices to make, and your taste’s fingerprints are what define you as an artist, despite the emergence of AI.
This whole post is a proof of concept, not the final shape AI-Graft will take. You could add videos, VR experiences, sculpture with 3d printing… who knows?
I hope to see some of you take a look at what I’m building for feedback, and so you can get a deeper understanding of it. It is still only a scaffolding for now, though, so expect inconsistencies. I’ve started adding images only 3 days back, the music is still added on the regular, and I’m rewriting the whole text itself, so there is still a lot of incomplete and it requires some editing for sure post rewrite.
So… welcome to the future of art, ladies and gents! A future where you can compete with yesteryear’s multimillion-dollar projects alone in your cave. Build Cathedrals out of it if you dare make it Baroque!