Using the Mind as a Camera

[Michael Erlewine]

I can’t help but bring to mind the quote from the song from “The Ballad of a Thin Man’ by Dylan “Because something is happening here, but you don't know what it is Do you, Mr. Jones?”

In this case, that Dylan quote would include me. I hitchhiked (traveled with) Bob Dylan back in 1961 and helped him put on early concerts. I don’t talk about my music carrier on this blog. Here I want to share something that should interest photographers.

I was not at first aware of this particular sea change I mention here. It’s not unusual that great changes happen in our midst, all around us, and we only eventually find out about them as they emerge and become popular enough to be seen.

This is happening right now with artificial intelligence as it relates to taking dictation from us in verbal form and using our mind to create an image that we perhaps find usable or even outstanding.

The online AI software I am using is called Midjourney (there are others), and I have been creating images with Midjourney for a while now. What’s amazing is that even over the short time I have been creating Midjourney images, they have gone through a few major upgrades, and I mean major. This has just happened with what is called Midjourney Version 4. This new version dropped yesterday; I believe.

Version 4 is such a major change, and it is shocking to see how Midjourney lurches forward into the future of image creation, where anyone, I mean people like us who are photographers, can also create art by simply writing out what kind of image we are looking for. I do this almost every day now.

I have been publishing a blog on Facebook every day since 2007, which is quite a long time. I have 5000 Facebook Friends” and another 5000 followers, and about 1000 who want to be friended, but I have run out of room

Midjourney AI allows me to enter any series of words to describe an image I am looking for and then does its best to produce images, offering four images for each entry, which I then can accept or ask Midjourney for variations for a particular image.

Perhaps you might like to see some of the sentences I feed into Midjourney V4 and the images it produces. These are not something I intend to use, other than for illustration here, yet they illustrate the advance forward this version provides.

Since on my blog, I sometimes post about my music career, astrological careerr, music database (AllMusic.com), movie database (AllMovie.com), and of course my interest in photography. I post a lot of my nature photos, some of which I post here, but I also have been finding and licensing images for years. Since I found Midjourney, I no longer license any images, but just create them using my mind  as the camera and Midjourney AI. Works wonderfully.

Just to be clear, this is all online. All I do is type in, as best I can, words or a sentence that I want to have an image created. An example would be, using dharma images since I have run a meditation center since the 1980s and often post on mind training:

 “Buddhist monk discovers the true nature of the mind”

 “In deepest meditation a Buddhist monk rests in the true nature of the mind.”

 “Buddhist monk on a cushion in a dark room with a candle rest in the true nature of the mind”

Now, these are not simple concepts, of course, but I hope that these few examples will give you an idea as to what can be done. I simply feed the program a sentence describing what I want my images to look like. Midjourney then produces four examples. I can choose to save one or more of them or I can choose one of the four images and ask for variations on that image. This process of trial and error can go on for many iterations, until I see one image that expresses well enough what I am looking for. I then use that for my daily Facebook blog.

The takeaway is that you and I may be good photographers but may not have the artistic chops to draw an image like these, but we do have the inner mental vision (and composition sense) to know when an image comes close enough to satisfy us. In that sense, each person can produce meaningful art, both for their own enjoyment, but also to show others what and how we feel inside. That is powerful, IMO. And it will enable each of us to express our inner artist.

Midjourney was originally launched for public Beta on July 12, 2022, which was not that long ago. And since then, there have been four versions, V4 being the current version that was just launched days ago.

I have watched Midjourney AI grow through four versions, each time getting more and more realistic. It is only a skip and a jump until these AI image programs allow us to seed them with our mind and produce useful images that require a skill all their own, IMO.

[Jack Dahlgren]

Michael,

This is a very interesting time isn't it!

There are a few points about it that I find very interesting.

The first is that the bridge between words and images is emerging from the fog. Through history we had sayings such as "a picture is worth a thousand words" which show the difference between what we see/imagine and our vocabulary to describe it. Training a machine on millions or billions of examples to learn the relationship between an image and the words which were used to label it allows it to determine the patterns which underlie both the words and the images. And then make new images using this new vocabulary. It is indeed a new vocabulary where words have meanings which may not have been what was originally meant. Artist names in the set of words we feed the machine may conjure up images which embody the style of that artist's work, or it may produce a portrait of that artist if there are many images of the artist in the training set. I think we are just at the beginning here of something that is moving very rapidly. I think it will move beyond supplying lists of words to embodying perhaps a new grammar of images. This grammar may be textual, or may be sliders or knobs on an interface, just as we find a reverb knob on a guitar amplifier. I also see this extending from images to multiple dimensions and from descriptions to actions.

The other thing I find interesting about this is what the source of this is. As I mentioned, this is the result of a machine deriving relationships between billions of images and words which have been found in our human world and as such it embodies the characteristics and assumptions and biases of the information which went into the model. Some we are unlikely to be able to tease these out on our own. It is almost like a collective unconscious. I believe the next step is to move from the unconscious to the conscious training of relationships. These may be done by artists who want to build and share their model/style with others, or by those who want to explore, expose, and perhaps eliminate biases, and maybe by those who seek to understand the vocabulary and grammar of the visual in other languages and cultures. Finally, there will be those who want to move beyond the unconscious memories of what has already been done and make something new. The last is difficult. Just as we gravitate towards existing languages for their stability and universality, we have similar tendencies in the world of images. Going beyond those shared agreements (even if not universal) may make the work unintelligible and it will perhaps be see as being devoid of "beauty".

[Michael Erlewine]

Jack,

for my own two-cents, I believe that for people like me, aside from the visual empowerment, it will be the need to use this new technology on a daily basis, as I am, to accomplish what needs to be accomplished. That provides the iteration and practice needed, not so much making art, as for just illustrating. Different strokes for different folks. I'm not so interested in this as art as I am in this being a utility.       

[Robert Camfield]

"...I think we are just at the beginning here of something that is moving very rapidly. I think it will move beyond supplying lists of words to embodying perhaps a new grammar of images. This grammar may be textual, or may be sliders or knobs on an interface, just as we find a reverb knob on a guitar amplifier. I also see this extending from images to multiple dimensions and from descriptions to actions.

The other thing I find interesting about this is what the source of this is. As I mentioned, this is the result of a machine deriving relationships between billions of images and words which have been found in our human world and as such it embodies the characteristics and assumptions and biases of the information which went into the model. Some we are unlikely to be able to tease these out on our own. It is almost like a collective unconscious. I believe the next step is to move from the unconscious to the conscious training of relationships. These may be done by artists who want to build and share their model/style with others, or by those who want to explore, expose, and perhaps eliminate biases, and maybe by those who seek to understand the vocabulary and grammar of the visual in other languages and cultures. Finally, there will be those who want to move beyond the unconscious memories of what has already been done and make something new. The last is difficult. Just as we gravitate towards existing languages for their stability and universality, we have similar tendencies in the world of images. Going beyond those shared agreements (even if not universal) may make the work unintelligible and it will perhaps be see as being devoid of "beauty"."

Jack,

Your thoughts have real insight, touching on the potential for AI capability to enable cross-media artistic content, as Michael demonstrates. Could AI reach beyond narrative to visual art forms? Just imagine: a composition akin to V. Williams' The Lark Ascending...a derivative of 19th century poetry as I recall...written by AI.

The sheer size of contemporary AI work enabled by supercomputers is astounding. These days, the algorithms of neural networks can apparently search through >100 billion data records, ferreting out relationships to carry out remarkably complicated tasks. Recently, a science writer for the Economist collaborated with the AI lab at Stanford: constructing a share of a complex narrative on a science topic using AI. The result was remarkably lucid.

[ColinM]

When Michael first posted this, I was intruiged but didn't follow it up.

This week, I have seen people post pictures online that I initially liked.
See below and the post which follows

[ColinM]

The original poster didn't give his source, so I did a Google image search on the file.
The results yielded no direct match, but lots of slightly similar images.

The majority were actually paintings, using the original Vermeer and Picasso influences to try producing their own "original" art.

When I queried this with the original poster, he then admitted he had used an AI package, feedng in the Vermeer work as part of the source data.
I looked at it again and my initial enthusiasm waned

[ColinM]

Today, I've seen two further posts by the same person (see below)
By now, they are just of passing interest as no matter how apparently skillful the results are, they seem to have lost something.

And it reminded me of apparently spectacular photos I see shared on Facebook that some of us can see are the results of subtle (or not so believable) use of Photoshop.
And yet many people are bowled over by them and may give up, because they think they'll never be able to achieve the same results with their own phone or camera.....

What do you reckon?