Written by Some time back you may recall I published some member content regarding the mind’s “internal monologue”. It had come up from a caller into the public show that while I was without Internet access this had been a big discussion on social media that many people do not actually conceptualize the majority of their thoughts as words. This seemed quite bizarre to me because my world is so verbally centered that this could be misconstrued as to lack thoughts.
On Monday I noted a bit of writer’s block. This drives me insane. I sit to write and nothing happens. To me this is the functional equivalent of erectile dysfunction.
To make productive use of my time in the absence of my verbal powers, I decided to delve back into my technology projects. Early in this production, those of you who have been around awhile will recall my fascination with the advancements in Artificial Intelligence which had taken place in my absence. When I landed in December of 2022, I had never seen an AI image generator, and to see them produce such stunning visual productions I was shocked. Discovering that this was available to anybody with a few hundred dollars, I was determined to harness the power of this for myself.
It was of particular interest to me specifically because of the aforementioned verbal centrism. I cannot draw to save my life. Even my handwriting is terrible. I have no idea what makes a good photo a good photo. My show art prior to my 2020 abduction was comprised largely of images from other sources with my font and branding across the bottom.
Even outside of art, when I was in the technology industry, visuals proved difficult. I worked as a project manager for a company that ran cable and built datacenters. The measure distance by eye is a preposterous concept for me. I draw a blank when asked to estimate crowd size at a distance. Cable installations to me are functional things, their vanity is largely lost on me. It was not uncommon that if I had to fill in for a truck crew to do an installation, somebody would be by later to make my work more presentable.
The idea that I could tell a computer in text what I wanted to see, and then it appear, could fairly be described as life altering to me. Images are powerful. The old “a picture is worth a thousand words” cliche is a popular one for good cause, and I do not compose a thousand words so quickly.
I have had occasion in recent months to converse with more than one visual artist. I have developed a new appreciation for their craft as of late, and I must strive to keep this in the category of admiration, lest it slip into envy. I found this an interesting thing to do since now my theory of mind for them had been altered. I imagine they think in image, and I in word. I communicate to them in speech, and I imagine a process of translation happening in their mind. Accounting for it, I attempt to describe a visual, a weakness of mine, and a talent I must refine also for the AI projects.
Harnessing the power of the AI image generators proved a frustrating effort. When one first happens upon them, it appears they are fountains of limitless creativity, but for a man in my position the limits are eventually discovered. I was happy to find, notably, that my public infamy allowed me to create likenesses of myself by use of my name. Unfortunately, it often things I am the fat guy in my slanderous Wikipedia photo, and mixes my image with that of a comic book author of the same name. Trying to tell it “a lean, muscular, Christopher Cantwell” makes of me a cartoon character, and telling the generators that I am the podcast host often does little more than place headphones on me.
Political correctness has hardly been gentle upon the most prominent AI developers. MidJourney has an ever growing class of nonsense they simply will not permit you to request, despite my payment for the service. Almost all of these platforms have something in their terms of service which will have you banned, often without refund, for attempting to circumvent their censorship.
I came across a self hosted system called Easy Diffusion. A web interface for a locally hosted Stable Diffusion server. It took time to generate images with my RTX 3060 video card, and the images it produced were unimpressive, even as I loaded different models into the system. I reached out to financial supporters seeking to upgrade the video card. I did, and it produced unimpressive images much faster.
I was furious, and I back burnered the project. It has sat there largely untouched for several months.
With my verbal talents in deficit, I went through my large pile of incomplete tech projects, and decided it was time to get my visuals up to snuff.
Fortunately, my mind still works, though it may be lacking for eloquence or inspiration in the moment. In fact, my being uninspired may have helped me to focus on the duller aspects of the AI work.
When I am feeling inspired, the drive is to express these things through the page or the microphone. I must allow my mind to run freely and focus on technical tasks can be challenging. To sort through data, to read a technical manual, to copy, paste, resize, drag, and drop, the mundane tasks of a computer user, to toil at a command line, I become frustrated and scream internally that my genius must not be burdened with that which a non-artist could be trained to do.
Deprived of my verbal powers, the dry technical things are less boring. They find a sort of inspiration of their own. They put wins on the board, and thereby restore my confidence.
I have long wanted to “train” an AI image generator to produce more reliable likenesses of your humble correspondent, so as to use for memes and show art. These last few days I have made remarkable progress in this.
In doing so I have finally gotten down how to produce a LoRA, or Low Rank Adaptation model, to supersede the larger data sets that comprise the generators with carefully selected images associated with words.
Thus, my local machine now places Christopher Cantwell wherever I say, and you cannot mistake him for anyone else.
Another thing I found frustrating about the image generators was their refusal to do anything that could be interpreted as violent. Notably “A man punching through a brick wall” proved impossible for me to do on my local computer, though MidJourney was semi-capable. Other scenarios proved impossible for both, such as when I wanted an image of a Judge’s gavel smashing through a courthouse.
I trained a LoRA last night to make punching through a brick wall as easy as saying so.
My mind runs wild with the possibilities. Whenever I find a concept the system struggles with, I can feed it images to train it now. Each time I do that, the system then produces images, which I can use to train it further on its own productions.
Give the machine a man. Give the machine an object. Give the machine an action. Have the man perform the action with the object in one style, then another, then another, repeat, repeat, repeat. Feed the images back in to train the system on its greatest accomplishments. Refine, revise, undo, try again.
I am inspired once more, though not in the happiest of ways.
I do not think of myself as old. Giving the computer images of me taken recently, it thinks otherwise. The lines around my eyes. My nose exaggerated.
I try to fix this in a way I think very clever. I blend my face with those of more handsome men, and feed it back into the system. I think I am very clever, and I put several hours into the task.
The success of the effort is limited. I’m still old. My nose still stands out. But now I look fake.
Punishment for my treachery, and a reminder that time is a cruel master, and a harsh judge.
I’ll show some of this to you on video tonight. I’ve been fascinated by it and I hope you’ll find some entertainment value in it. The possibilities are seemingly endless and I have some hope that it will prove as inspiring as it is useful.
I’ll have much more to say about this when we meet tonight as we do every Wednesday at 9:30pm US Eastern for our SurrealPolitiks Members Only Video Chat.
Recall that we are now hosting the chats on Discord!
You can link your SurrealPolitiks Member account to your Discord account using the links provided either on the Member Chat Page https://surrealpolitiks.com/member-chat/ or on the Your Profile page on SurrealPolitiks.com https://surrealpolitiks.com/membership-account/your-profile/. This integrates a Discord Bot with our membership software to control access to member areas.
If you are having trouble connecting your account, there is a link to the Discord Server on the member chat page and you can join the server without your account being linked. From there someone can try to help you get things sorted out. If you don’t have a Discord account, you can create one on Discord.com. You can get the app for Windows, Linux, Mac, iOS, or Android fairly easily by searching their respective app stores or getting it directly from Discord.com.
If you’re not a member yet, you can become one at https://SurrealPolitiks.com/join for just $6.70/month if you use promo code agenda33 at checkout.
Once you’re a member, you can join us every Wednesday at 9:30pm Eastern on our Discord Server, and you can find these instructions on our Member Chat here https://surrealpolitiks.com/member-chat/.
If you miss the live show, you can always catch the replay in the member content section here https://surrealpolitiks.com/category/membersonly/.
Those of you with Odysee content subscriptions can find the Odysee stream here https://odysee.com/@ChrisCantwell:9/spmc20240320:a