vessels
As ai gets more accurate it is becoming more difficult to create vessels with asymmetry. Could this be because of historical and contemporary ideals of art and the collection of images as sources? Because so many find symmetry pleasing? Whatever the reason, it is hard to break through. Some models are especially challenging, particularly in visualizing the outer silhouette.
Images generated with Leonardo.ai
A rewarding part of playing with ai has been collaborating with exactly.ai and getting to articulate why I enjoy using their image generator. Below is a series of my images cross-posted on Instagram and my answer to “how do you use exactly.ai?”
Images generated with exactly.ai
I have always had a fascination with archaeology and the sea and I continue to circle back around these topics. For quite some time I mainly worked in printmaking and multimedia, but I’ve recently refocused more around ceramics and drawing. This has led down an interesting path both in terms of concept iteration and in what I’m producing.
I started thinking about ceramic vessels— if sat in the sea for a long time what would accumulate on them and how could they mutate and take on a life of their own. I’ve continued to follow this line of thought to see where it will bring me both conceptually and visually.
Bing’s Dalle-2 was a fun place to play out some of these ideas and concepts. Here are some initial iterations of vessels with the eye as an inherent part of a seashell, sometimes manifesting as a pupil (hole) or an operculum of sorts. Though they are symmetrical they are not too uniform so there is an uncanny handmade quality to them. The ceramic texture are spot on as are the imperfections.
Images generated with Bing.ai / Dalle-2
Meta ai’s Llama-3 feels pretty much like Bing’s Dalle-3. Images generated with the same prompt seem to live in the same family. Here are more pepto pink images to compare to those from Bing in the previous post. Though these are supposed to be ceramic they look more like glass to me.
Images generated with Meta.ai
Playing with Bing’s image generator I found Dalle-2 to be my favorite version. The images I started generating in early August had personality! I particularly liked the mix of realism and quirky ai oddness. Bing has moved on to Dalle-3, which I find to be too over the top. Every image generated is completely covered in stuff from top to bottom like it’s trying too hard and there is no nuance to these images. See the difference between the photos below using the same prompt, Dalle-2 first, then Dalle-3. Dalle-2 is no longer available for free unfortunately. I wish it was, I miss it!
Images generated with bing.ai / Dalle-2
Image generated with bing.ai / Dalle-3
So it may seem easy, but having images in your mind is a far cry from what you may get with text to image prompts. It’s an art of sorts to push image makers to create something that is “successful.” The process is similar to photography in that many, many images need to be generated and then the best is selected through an editing process. Image generation has the added step of writing, tweaking and re-editing the prompt many times to try to veer the generator in the desired direction. As I continue to generate images, prompt writing is a constant challenge. The prompt is a living, always changing and evolving part of my ai project creative process now.
Images generated with stable diffusion 2.1
As I was trying different image generators I discovered durer.ai, which has since become exactly.ai. Its models at the time provided a varied and complex range of more mature art styles, and for me an interesting photo-realistic aesthetic. This generator has continued to pique my interest as it has matured.
Images generated with durer.ai (now exactly.ai)
Here are a few results using a “cute creature generator” from deep ai. The anthropomorphism angle is always an interesting one.
Images generated with deep.ai
My fascination with the sea led me to initial ideas about Atlantis and what a ceramic world might look like under the sea. This notion quickly gave way to a more focused look at the ceramics itself, and became more about the piece rather than its surroundings.
Images generated with dream.ai
With image generation there always seems to be a push-pull with super detailed/ overdone vs. minimal, depending on the model and the prompts. Here are some super-detailed extra-slick images. Seems the more ai develops the more slick and less error-prone images are becoming, which is distressing in terms of real/ fake. I prefer elements of the unexpected to help push my ideas.
image generated with Leonardo.ai
Images generated with Leonardo.ai