I have yet to embrace or experiment with AI image generators but I can see the potential though. Instead of waiting months for the right weather to make a stormy seascape –as I did with this image above — I could generate one with a text prompt using generative AI. I could even do a seascape series that would include various objects and different photographic period styles without leaving the computer.
I am tempted to play around with the groundbreaking Midjourney as its powerful algorithm creates is high quality and photorealistic images with resolution up to 2048×1280 pixels. Tempted because it means not having to pay for old film cameras to be repaired at great expense, or paying for film and development by a lab and software to produce images, or needing to buy expensive digital photography equipment to transition from 5×4 film to medium format digital. With generative AI you don’t need any fancy tech gear or expensive software to produce images and that’s rather appealing.
My hesitancy with embracing text-to-image AI art is that the image generators are still in beta mode and, with the exception of Stable Diffusion (which is open source) they are not free. So they’re currently harder to just play around with, in the sense of doing a bit of trial and error to learn how its AI tools work inside and out. Midjourney’s current limitations emerge from the program being trained using pre-existing real-world data, and so it is challenging for this AI model to create something completely novel or an image that is above and beyond what is conceivable in the real world.
Presumably, the technology underpinning the AI industry will continue to improve, thereby enhancing Midjourney’s image generation capabilities. With the V6 version you can now add text into the image. The software companies like Adobe or DXO are starting to provide a whole array of new generative “AI” tools, which can, if the user chooses, be used to modify photographic images; and they’ll also provide ever more advanced tools to generate non-photographic images for those who choose to follow that route.
Update
Is it time to dip my toes into the AI image generator waters?
I won’t be using the generative AI darling Stable Diffusion text-to-image model as the UK based Stability AI, which runs Stable Diffusion, is running out of cash. It has incurred big debts it can’t repay, is facing multiple copyright infringement cases brought by Getty and other artists, and is looking for a buyer. The bust after the hype of the AI boom has begun, and it is caused by a looming cash crunch of the state of the art, frontier AI models from the high cost of both talent and the GPU infrastructure for the computer power needed and the meager revenue opportunities currently available. That does not look good for the open source models.
Midjourney is the most photogenic. I realize that though everything that initially I do will be based on something that already exists (ie., older images), this image generation is not a mashing or stitching existing images together to create something new. Machine learning is built into the technology and so AI as an form of generating image can move forward when we stop imitating, and use the technology on its own merits, with ideas and concepts unique to the medium. That requires lots of experimentation.
I reckon it will take me weeks to start understanding how the prompts work and create images with a faintly interesting aesthetic. There is no free access to dabble in Midjourney so as to get away from repeating existing imagery. I have to sign up to using Midjourney’s basic tools if I want to generate an seascape with AI with the basic plan being $US10 a month. This gives me 3.3 hours per month with options to top up. 3.3 hours per month is not much time to learn how its AI tools work inside and out.
No Comments