AI wins the best photo
- Thread starter DLD
- Start date
ahalpert
Major Contributor
Not exactly a photo. More like a digital painting printed on a canvas. Also worth mentioning, as the artist himself emphasizes, that the human behind the creation was instrumental in directing the program, curating the options it gave him, and tailoring the results. What we're seeing is the death of the technician, not the artist. People who earn their living by their technical expertise alone, rather than their creative vision, are at risk of losing their livelihoods
I have used DALL-E & Stable Diffusion. Pretty sure Midjourney was used by the designer for this competition. These are text > image generators. At this stage these almost always produce 'off' results to say the least, particularly faces (animal, fantastical creature or human). Things are progressing though. I think these things are going to absolutely eliminate some amount of illustrator/digital artist jobs & the stock imagery marketplaces in particular.
Stable Diffusion is open-source, is far less prohibitive of what images can be generated - and key - can be run locally with a powerful enough GPU. Their model is probably going to dominate. Other firms will use their codebase and offer specialized generators for particular clients & aesthetics (already are).
People claim this technology (which is fascinating btw) can create animations & video as well.
Ironically what is actually needed right now most when using these tools is good technical understanding of digital painting programs to get the output to be passable as a final image (professionally).
Last edited:
The AI generated artwork is pretty interesting and nowhere near as automatic people think. You have to go through a lot of iterations to find what you are looking for and even then some of the results can be pretty random. The below images were generated with the description "The Archangel Michael's victory over the great Dragon Lucifer, very detailed 8K oil painting"
Last edited:
Word association I assume. If you write Picasso or Dali etc it will emulate those styles. You can specify mediums like painted or illustrated or realistic, etc.
Midjourney gives you 4 different images, which you can then request more iterations of the ones you prefer, and so and so on.
Midjourney gives you 4 different images, which you can then request more iterations of the ones you prefer, and so and so on.
These machine learning programs are mining enormous online image databases as the underlying references the generator then produces through it's associations with the words you are prompting it with. These are datasets of hundreds of millions of images, next-generation systems will use billion plus datasets (along with resolution improvements I assume).
As Ag mentioned, the rendering of the style is based on the specific prompts you supply. You can also supply your own images (the most interesting option I am currently exploring).
Inevitably these engines will be integrated into programs like Photoshop in a more productive way. I think this is one technology that's effect and development is going to be far more rapid & consequential than other automated, easily accessible 'creative' technologies. Perhaps a fad that will die off quickly, but a culture of 'create' your own 'art' instead of consuming other people's (outside of in-group) is a likely big development with something like this. Also the potential for business clients to use a program like Stable Diffusion with a much smaller staff for their graphic needs and turn away from the current options.
As Ag mentioned, the rendering of the style is based on the specific prompts you supply. You can also supply your own images (the most interesting option I am currently exploring).
Inevitably these engines will be integrated into programs like Photoshop in a more productive way. I think this is one technology that's effect and development is going to be far more rapid & consequential than other automated, easily accessible 'creative' technologies. Perhaps a fad that will die off quickly, but a culture of 'create' your own 'art' instead of consuming other people's (outside of in-group) is a likely big development with something like this. Also the potential for business clients to use a program like Stable Diffusion with a much smaller staff for their graphic needs and turn away from the current options.
With Stable Diffusion, you may never believe what you see online again | Ars Technica
Ars doing an excellent article on technology as usual
Ars doing an excellent article on technology as usual
very interested to hear what folks think in response to this part of the article:
"ISMs are also a dramatic form of image compression: Stable Diffusion takes hundreds of millions of images and squeezes knowledge about them into a 4.2GB weights file. With the correct seed and settings, certain generated images can be reproduced deterministically. One could imagine using a variation of this technology in the future to compress, say, an 8K feature film into a few megabytes of text. Once that's the case, anyone could compose their own feature films that way as well. The implications of this technology are only just beginning to be explored, so it may take us in wild new directions we can't foresee at the moment."
With Stable Diffusion, you may never believe what you see online again | Ars Technica
"ISMs are also a dramatic form of image compression: Stable Diffusion takes hundreds of millions of images and squeezes knowledge about them into a 4.2GB weights file. With the correct seed and settings, certain generated images can be reproduced deterministically. One could imagine using a variation of this technology in the future to compress, say, an 8K feature film into a few megabytes of text. Once that's the case, anyone could compose their own feature films that way as well. The implications of this technology are only just beginning to be explored, so it may take us in wild new directions we can't foresee at the moment."
With Stable Diffusion, you may never believe what you see online again | Ars Technica