Mostly Human: A surreal trip of man vs. machine where Gen AI meets fine art by Christopher Smallfield
Christoher Smallfield aka blckbox is a visual effect expert and generative ai artist. He is the creator of the generative AI art book Mostly Human. We met Chris to learn more about his project, his work process and his thoughts on the evolution of Generative AI art and a possible renaissance of a digital neo-surrealism.
© Christopher Smallfield, Mostly Human
Who are you? What is your professional background and what motivated you to create this specific project?
Hi! I'm Chris and I'm a Visual Effects Supervisor for Film and TV shows. Most recently I worked on the Disney+ series' The-Hulk, Ms.Marvel and Loki. I'm doing a lot of work for Marvel. My background is in the visual arts both traditional techniques such as drawing and painting and digital ones such as 3D and VFX. For the last two years however I have been experimenting with Generative AI Art which is the topic of my new book. I was motivated to make a book as a kind of future historical document now AI art has exploded in the last several months, it's been a firehose of imagery blasted onto social media and the internet and all nuance gets buried beneath it. A book is static and immune to the algorithms that push only the absolute latest things into view.
© Christopher Smallfield, Mostly Human
What is the meaning or the narrative behind your book title „Mostly Human“?
Like the title, the artwork within is generated by a human, the trained models that were used were made by humans and the source material on which the models were trained is also human, but the result is not entirely human. I wanted to share the feeling I had when I saw the first image created through AI generation instead of procedural methods. It had something magical in it. Like all art, it's an interpretation of an idea and this image I was seeing of a lighthouse was simultaneously familiar and alien, The brushstrokes were organic, the rendering beautiful but then obvious things were warped, missing or simply wrong, in a way a human likely would not choose. I think my favorite thing about generative AI art is that you don't get exactly what you expect, it is a happy accident machine.
How would you describe the visual style of your generative ai images? Are you inspired by specific artists? What is your favorite software and why?
I would say that my style, as it has been in traditional art, is ever-evolving. I like to explore both in terms of ideas and in terms of the limits of the medium. I would consider AI art more a medium than a tool as it's often described. In this book my style is centered around surrealism and expressionism. At the time I made the work, Disco Diffusion was the most advanced tool and the tool I used. I still think that in terms of style it's a very powerful tool, in terms of coherence of images it doesn't stand a chance against Stable Diffusion or MidJourney.
Part of what I wanted to express with the book is that the idiosyncrasies of each tool are things to embrace and lean into. I saw a lot of frustration in artists that were trying to achieve total realism with Disco Diffusion. Its strength is more in its alien poetry, its ability to take your idea, your references, your inspirations and and construct a compelling image, or not. Plenty of times the results were awful. I'm using MidJourney and Stable Diffusion more lately and plan to continue making books as these art forms evolve and hopefully as I do too. I am inspired by many artists and styles, for the cover James Jean was a reference. The image looks nothing like something he would make but there are aspects to the lines and colors that are reminiscent. Trying to copy your inspirations is a bad goal in my opinion. Remixing them, reinterpreting them, then turning the results upside down and inside out can be wonderful though. Generative AI is a perfect tool for such things.
© Christopher Smallfield, Mostly Human
How would you evaluate the importance of curation, knowledge and education for the creation of generative ai art? Is generative ai rather a tool or a co-creator for you?
Curation is the number one skill one needs to develop to have good collections of work with Generative AI. One of AI's superpowers is to instantly and endlessly iterate on an idea, that's an aspect that no human can compete with. Sometimes I've written a prompt that is almost resulting in what I'm wanting and any change I make to the prompt takes me further away so instead one can say, "Give me 1000 variations of this almost-correct prompt." and you can sort out the results. That can be harder than it sounds, I'm typically narrowing down the 5 images I want from a pool of 500. For me, the best way has been to do it in rounds. Filter out the clear failures first. Then pick all the pretty good images. After that it gets harder as you might still have 100 images that have elements your really like. I make up rules and try to trim it down that way, but if I find myself making an exception over and over for a certain image, then it stays in.
Rarely I take an almost-great image and manipulate or paint over parts of it, other artists do that a lot with great results, but for me personally it feel antithetical to the point of the tool. A background in visual arts seems to be an advantage. Creative people in general, writers, musicians also tend to make better images. I can usually guess if someone has a Pre-AI artistic background from the images they generate, but not always. For me, I don't consider it a co-creator as all current implementations don't "think" in a human way and I think that assigns a kind of agency to it that it doesn't have. I consider it a medium more than a tool. It isn't a tool for making a photograph, that would be a camera. It isn't a tool for making a painting, that would be brush or palette knife. It's software that can make images that are photographic or painterly but they are always something I would classify as generated. I think it's important to be open and honest about the method's one uses. I'm not a fan of people submitting AI images to photography contests in the same way as it would be odd to submit a photograph to a painting competition. If it's to have any value as art it must stand on it's own, be what it is. That's why I would say that Generative AI art is the medium and something specific like MidJourney would be a tool.
© Christopher Smallfield, Mostly Human
In your opinion, are there currently any limitations at all to the creative creation of art via the current generative ai software solutions? If so, can we expect even more fantastic results in the future? How would you predict the future of generative ai?
There are clear limitations to current tools, but they are largely technical and the obvious ones are resolved in weeks and months rather than in years like other digital image making methods. The existing software today is excellent at making still images but still weak in making video and very weak in making 3D data such as models, textures or shading. In a year I imagine video will work as well as still images work now. 6 months after, 3D data will be the same. I think prompting will also continue to evolve and will become more discussion-like as we start to see with ChatGPT and other Large Language Models. You might start simple with your prompt and then get questions back to help the model understand your intent better. Once an image is generated it will be able to be refined conversationally as well. A bit more blue. Can you make it taller and shinier? I think that will be the case with video as well. We'll finally have definitive answers about whether our ideas are good or perhaps very generic or even bad, because executing them will be possible to a high degree.
© Christopher Smallfield, Mostly Human
With the rise of AI, do you think we could witness a renaissance of something like digital neo-surrealism that will also turn into a historic categorie of modern art or is generative ai art rather unclassifibale from your point of view?
I talk about this a bit in the book, that we are seeing trends in AI art now that cycle up and down then fall into obscurity in a matter or days. At the time of writing this, we just passed the moment where lots of people are making "Wes Anderson Versions" of popular movies and now people are making QR codes that look like landscapes and characters, By the time someone reads this that will likely feel like ancient news. It's like if the PreRaphealites started their movement and gave up on it after about a week and a half. I think that's a combination of social media's design, to keep one scrolling and refreshing endlessly and AI softwares very fast development cycle. It's another reason I wanted to make a book, to be a static bookmark in time. I don't think it's necessarily good or bad that we have these rapid trend cycles beause I do believe that bigger, higher quality movements will emerge. There are a handful of artists that have cultivated consistent styles and bodies of work despite the changing tools, I'm extremely impressed by that. Douggy Pledger has translated his creepy and darkly humorous style of art to AI and is just releasing his second book, Midnite on Mars (Steve Glashier) has made a whole alien world in soft pastels and Izanami is consistently making stunning sci-fi scenes with a lot of feeling. There are a lot more as well.
Creator Christopher Smallfield AKA blckbox
Want to know more about Mostly Human by Christopher Smallfield? Check-out this link: https://blckbox.tv/mostly-human
Connect on Instagram @blckbox.tv https://www.instagram.com/blckbox.tv/ or on LinkedIn https://www.linkedin.com/in/spacemanchris/
Mostly Human is available on Amazon: https://www.amazon.com/dp/B0C4N42L9G
