NikonGear'23
Gear Talk => Processing & Publication => Topic started by: Michael Erlewine on April 06, 2023, 03:26:43
-
AI Midjourney Imitates My Photographic Style
Hold onto your hats, because here is an innovation in AI technology that took me by surprise.
As a system programmer by trade, learning some of the new AI programming languages is easy, since they are just a prompting language to the AI software.
I have been working with Midjourney AI since it first dropped last summer (2022) and liking it. I have never spent any time complaining about what we as photographers will lose, but instead what I wanted was to learn how to use this new interface for my purposes as an illustrator. After all I have been writing blogs to around 11,000 people each day, and doing this for over 15 years.
And while those I know around me who are interesting in exploring Midjourney and creating ‘Art’, I never blinked an eye in that direction, and immediately set about illustrating my blogs, and cancelled my subscriptions to various stock image sites within hours of starting Midjourney.
I have not licensed one single image from photo stock companies since then. All that I wrote above is what other folks like me are doing all over the world. Yet hold on. Here comes what I did not expect to come down the pike, and, as mentioned, it took me by surprise. I didn’t see it coming.
Midjourney just added a new command called “/Describe” which allows me to input my own art, and in this article, I will used my own stacked photographs of flowers, which some of you reading this are perhaps familiar with, my style of close-up flower images.
And all I can say is “Whoa!”
In a matter of seconds, Midjourney accepted an image, one of my stacked flower images, and output to me four numbered steps in breaking down my style, each step containing a paragraph or two describing how my style works.
And then, to top it off, it Midjourney shows me four examples of the “Michael Erlewine” photo style, and I was stunned. These photos look like my stuff, yet are not copies!
Now, I just did this minutes ago, and I can only imagine what will happen if I tweak these photo-descriptions, which I probably will. Many of my family and friends will not be able to tell the difference between my own photos and these AI productions, using my style.
I just wanted to share this information at once with the group here, so that we can get some feedback here perhaps have some meaningful discussions.
These are not produced with any of my cameras, but only by Midjourney AI software. My sense is the AI is moving ahead not just exponentially, but at warp speed. Your thoughts?
-
Oh, AI is trying to mimic your signature style of photography... and that fairly successfully...maybe except that these images suffer from harsh lighting, lack of pleasant moisture and compositional skills. :o :o :o
-
Far from similar to Michaels work, but getting there.
Anyhow where mankind faces redundancy!
-
Stunning, Beautiful. But they have some roughness that your sharp flowers does not exhibit. Could perhaps be tweaked.
/Eric
-
I made it clear that these photos are rough, without any tweaking on my part. I'm sure that working with these images will result in better and better images.
This post is just to point out what is happening in AI these days.
I am not looking to use this instead of photographing.
I have to look into how best to use this. The recursion of the technique is great and this is just a new tool.
To switch hats for a moment, here is an AI Midjourney image that I prompted.
-
To switch hats for a moment, here is an AI Midjourney image that I prompted.
Not being Luddite, but both the man and the dog look lifeless.
-
Not being Luddite, but both the man and the dog look lifeless.
Have at it, but watch what another year will bring, IMO. Look at what is coming up on our flank.
-
It really is interesting how something that has been worked on for a decade has reached a threshold where it has become so easy to achieve acceptable results and the pace of adoption is now so fast. What I see is that AI has distilled a language of imagery. It has derived from the images it has ingested, and the text that accompanies them (the labels) relationships between the pixels of the image and the words used in association with them. As humans, with enough familiarity of a body of work, it is not hard to conjure up an idea of an image given a text description -the view of the neighbor’s house in winter from Birna’s window through an obscure lens - for example. Where the AI amazes is in the breadth of it’s familiarity, and the ability to express the words as the image itself - in a matter of seconds.
The underlying mathematical model this is based on is a very large collection of images and accompanying text, an approach which is limited by the labels on the work. As the “language” associated with an image becomes more elaborate and which may extend beyond words to include geometry, composition temporality and other characteristics which don’t typically exist in photo captions we can only expect the AI to be more malleable and accurate in interpreting our wishes.
Michael’s revelation that an AI model at some level knows and can describe back his style is an indication that this intermediary language of images is real and useful. For me, the most interestin use of this language is for poetic images. It is still rough, and has no well defined rules of meter and rhyme, but I have no doubts that it will get there and in time we will have tools which allow us expressions which were difficult or impossible in the past, and even perhaps help us to understand what it is that we are expressing when we are working in the visual realm.
These models are dependent on what we feed into them, so for best results human photographers should continue to create images and descriptions which are worthy of being a basis of a new language.An AI trained on the images and words posted in these forums would be an interesting thing indeed.
-
For me, the most interesting use of this AI language is for poetic images.
I totally agree with Jack. I write quite elaborate blogs to around 12,000 people, usually on a daily basis, many on dharma and related spiritual topics and I have had to search far and wide to find images with enough 'juice' in them to help spark readers with the content I write. I used to search hundreds of thousands of images on stock photo sites looking for the few images that had some 'feel' emotionally, yet with Midjourney AI, I have as many 'touchy-feely' images as I could want. I see many folks using Ai graphics and calling it 'Art', yet that does not interest me at all. AI graphics, IMO, is best for illustrating, which is what I need. Here is an example if an AI image with 'feeling'.
-
At the risk of sounding like a zealot, what you describe Micheal is exactly where I think the value is, as an extension of the written language into the visual realm. Imagining them to be “neo-pictographs” is how I think of it. AI is just the stylus and burnt minerals by which we draw them. AI is also the translator we can use to query the meaning of those same pictographs. As someone who learns and thinks visually, the prospects are so intriguing.
-
AI Midjourney Imitates My Photographic Style
Hold onto your hats, because here is an innovation in AI technology that took me by surprise.
As a system programmer by trade, learning some of the new AI programming languages is easy, since they are just a prompting language to the AI software.
I have been working with Midjourney AI since it first dropped last summer (2022) and liking it. I have never spent any time complaining about what we as photographers will lose, but instead what I wanted was to learn how to use this new interface for my purposes as an illustrator. After all I have been writing blogs to around 11,000 people each day, and doing this for over 15 years.
And while those I know around me who are interesting in exploring Midjourney and creating ‘Art’, I never blinked an eye in that direction, and immediately set about illustrating my blogs, and cancelled my subscriptions to various stock image sites within hours of starting Midjourney.
I have not licensed one single image from photo stock companies since then. All that I wrote above is what other folks like me are doing all over the world. Yet hold on. Here comes what I did not expect to come down the pike, and, as mentioned, it took me by surprise. I didn’t see it coming.
Midjourney just added a new command called “/Describe” which allows me to input my own art, and in this article, I will used my own stacked photographs of flowers, which some of you reading this are perhaps familiar with, my style of close-up flower images.
And all I can say is “Whoa!”
In a matter of seconds, Midjourney accepted an image, one of my stacked flower images, and output to me four numbered steps in breaking down my style, each step containing a paragraph or two describing how my style works.
And then, to top it off, it Midjourney shows me four examples of the “Michael Erlewine” photo style, and I was stunned. These photos look like my stuff, yet are not copies!
Now, I just did this minutes ago, and I can only imagine what will happen if I tweak these photo-descriptions, which I probably will. Many of my family and friends will not be able to tell the difference between my own photos and these AI productions, using my style.
I just wanted to share this information at once with the group here, so that we can get some feedback here perhaps have some meaningful discussions.
These are not produced with any of my cameras, but only by Midjourney AI software. My sense is the AI is moving ahead not just exponentially, but at warp speed. Your thoughts?
I really like the first two photos. They look painterly, and as if an oil painter tried to make a painting of one of your stacked flower images.
I'm into photography to create art, and not simply replicate nature and the world around me. As such, I view this AI software as another tool that can combine with my experimental photography methods and gear to create unique artistic images.
-
That AI is Frightening...feeling Uncomfortable about it.
-
That AI is Frightening...feeling Uncomfortable about it.
Like any tool, we have to learn to use it. Look at guns.
-
That AI is Frightening...feeling Uncomfortable about it.
What do you find frightening and uncomfortable about it? Part of my job is talking to people about AI so I’d like to understand your perspective.
-
I made it clear that these photos are rough, without any tweaking on my part. I'm sure that working with these images will result in better and better images.
I'm sure you noticed what looks like some glasses earpiece floating mid-air... You probably didn't ask for it in your description... Yes, additional work is needed!
Ciao from Massimo
-
I'm sure you noticed what looks like some glasses earpiece floating mid-air... You probably didn't ask for it in your description... Yes, additional work is needed!
Ciao from Massimo
Artifacts of any kind of sampling algorithm seem unavoidable and have to be retouched. Focus stacking is replete with artifacts unless you are very skilled or lucky, IMO. I have done going on 10,000 Midjourney images and I don' (as with this one) bother to edit out artifacts with each one. Here I am just giving you an idea for illustrative purposes.
-
Reminds me of a comic book named "FONE", where in a sci-fi world a human and a humanoid are stranded onto a planet, only to find a supercomputer intent on publishing every combinatorially possible aggregation of words
Look here: https://loscomicsdemachete.blogspot.com/2016/09/fone-de-milo-manara.html (https://loscomicsdemachete.blogspot.com/2016/09/fone-de-milo-manara.html)
The original by Milo Manara is in Italian, here it's in Spanish...
The humanoid digs out a book that is telling their fate event after event, except there's glitches, where the human is riding a gigantic baby, or standing upon a pilflar (should be pillar, it disappears under the feet of the human), or the humanoid turns into a naked chick ::). Finally no more glitches until the end, which reads FONE (should be "FINE" in Italian, we could render it as "THE UND" in English)
AI as it is now aggregates existing content to create new content. The next step (challenging), where I would see real AI, is to create new content from scratch
In other words, give a monkey a typewriter, and maybe some day she'll write a poem
Ciao from Massimo
-
The next step (challenging), where I would see real AI, is to create new content from scratch
That is Second Life.