AI Text-To-Image Generation: Six months of progress in a week
08/31/2022
A week ago, Stable Diffusion was released.
It is an open source, next generation AI text-to-image generation tool that is causing a Cambrian explosion of apps that are going to redefine the way we participate in all visual creative fields.
A year ago, I was trying to figure out how to generate a video of images interpolated by Deep Dream that were synced to music.
I was inspired by this TikTok: https://www.tiktok.com/t/ZTRPrE57v/
It ended up being extremely complicated to set up and I gave up.
With the release of Stable Diffusion, I'm excited to start up this project again to collaborate with AI to create. Now I can use Deforum, which implements Stable Diffusion in a pre-made Python Notebook and input what I want without having to figure out the code myself: https://github.com/deforum/stable-diffusion
Learning these tools shortens the time it takes to get a creative idea out without having to hiring a big multi-media production or taking years to learn the technical and mechanical skills of a digital art program. Instead, I now have time to focus on music and I eventually will be able to combine my own music with AI tools that can generate a video. These tools were created by specialists (ML Engineers) so that generalists like me can realize their creative visions.
The progress at which these tools are going to improve becomes exponential with open-source. Permission-less development allows for experimentation at a mass scale.
And we'll start to see advances in fields that we probably can't imagine.
With AI as part of the production process, it shortens the process of getting an idea onto the screen and into a final production. One of the main themes of my life is utilizing technology to accomplish more with less and increasing the rate at which ideas can come into fruition in reality.
I don't see AI art as competing with artists, but rather collaborating with artists who have vision to create even better works of art and driving experimentation to push art forward and democratize access of these tools to a wider group of people who may not have had the opportunity to pursue this area of study.
I didn't go to art school, but I can learn art history by applying art styles of different periods of history to my ideas.
I can cycle through many different styles in the span of days and see where my creative and aesthetic tastes align.
Ultimately artists spend time to learn what their own aesthetic taste is through the experiencing and experimenting with many influences. Art is the result of remix, you need to gain experience with art styles to come up with your own unique twist and style. I view these AI tools as a way to generate new ideas quickly. Sometimes AI will be able to give us ideas that we may not consider since they are learning from the experiences from all of the artists from the training data.
I can try out multiple styles as I figure out my own style and aesthetic tastes.
I think one interesting comparison to what's happening in the AI text-to-image world would be how film and video makers on TikTok are learning at the speed of technology, they are speed running different film styles and getting live feedback from audiences from around the globe.
There's an explosion of talent in the video space as creators are getting exposed to every single film style and live experimenting using accessible consumer tools provided by TikTok and other app developers. It doesn't take much to create videos anymore except for a smartphone and a couple of apps and now creators can focus on the content and quality instead of having to spend years developing the technical expertise in a film school to get a particular aesthetic.
What would an AI-centric social network look like?
One where we could cycle through different types of AI collaboration and generated videos and images and see how we can advance and innovate. I'm in a few group chats and groups where people are experimenting with these new tools. It would be great to have a social network that would help with documenting the different methodologies and sharing collective knowledge.
The reason why I'm so passionate about this new technology is because I see its criticisms and they match the criticisms leveraged against other successful technologies: https://cdixon.org/2010/01/03/the-next-big-thing-will-start-out-looking-like-a-toy
Ultimately, art is a field where a lot artists have not been able to capture the value they produce.
And tools like Stable Diffusion are democratizing access to the ability to turn one's ideas into reality.
The future is one in which prompt wizards and AI-human collaborations will be a core part of the creation process in every single company and organization. The ability to articulate and communicate ideas is a core skillset for the future. And now instead of just interacting with humans, we will develop creative relationships with AI tools to help us realize our dreams.
I think the community around these tools is going to accelerate our creativity as we become more proficient in turning our ideas into reality.
The explosion of creativity after advancement of technology is the true driver of progress.
The last meaningful hardware technology advancement was the iPhone, but I'm excited to see how software tools like AI will help us move forward in the creative fields. I'm writing these essays because I want to figure out my why - why am I so excited about these tools? What do I want to create? What is possible now that I have this access and knowledge of these tools?
When I think about the future and what I may be doing, I don't doubt that interacting with AI tools like GPT-3 and Stable Diffusion will be as core to my workflow as emailing or messaging coworkers. I will now also have AI coworkers to interact with.
Personally, one of my favorite parts about this post is how you are willing to experiment with these emerging technologies regardless of how competent others might be in this arena, utilizing the crosspollination of interests and skills you've acquired over of the years.
I see these blog posts as deeper inquiries and excavations - mementos to look back on - and also leaving breadcrumbs for others to dip their toes in. It is fascinating to think about how intelligent algorithms can spark new ideas that we may not have considered to complement our personal imaginations.
Both documentive (sp?) and informative my dude!