Video has been part of the federation since 2023, and it never became important.
I am including this entry for exactly that reason. A ledger in which every category turns out to matter would not be a record of anything. It would be an advertisement.
The one that stayed bad
My first real experience with AI video generation was Kling, around September or October of 2023.
By then I was used to what the other systems could do. I could type sunrise, Sullivan’s Island and get back a genuinely useful image. I could ask for an article on the Reality Equation and, given enough context, get something at a professional level.
Video felt entirely different.
The early systems were text-to-video or image-to-video, and a “long” video was five or ten seconds. The quality was bad.
I tried virtually everything available. Kling. The other early generators. Talking-head systems. Avatar approaches where a synthetic presenter reads a script. Short clips, longer clips, image animation. I kept subscriptions running because I wanted to understand what was happening, and I was never particularly impressed by what came out.
Video was the first category where I could clearly see the promise and did not think the artifact was any good.
What actually improved it
The first real change for me came with Google’s NotebookLM, and what made it interesting was not that the pictures suddenly got beautiful.
It was the generation model.
I could hand it an article, a document or a link, press a button for a video overview, and get back something measured in minutes rather than seconds. I did not have to build scenes individually. I did not have to script transitions. I did not have to assemble anything in an editor.
I gave it the source material and let it generate the video.
That was much closer to the relationship I already had with every other generative system, and it is the only reason video became usable to me at all.
Why 12,000 of them exist
I began putting some of these on YouTube — not because I thought they were good, but because YouTube was a convenient way to preserve them as artifacts.
That distinction matters to this entire ledger, so I will state it plainly:
An artifact is evidence that something was generated and existed in the world.
It does not mean I believe it was excellent.
Since 2023 the federation has generated roughly 12,000 videos: very short clips, longer generated pieces, experiments, talking-head videos, shorts, article-based videos and a good deal else. I experimented with Sora. I experimented extensively with Google’s VEO. I tried essentially every meaningful stage of the technology as it arrived.
Even now, in August 2026, video is the category where I remain least impressed by the distance between what the technology promises and what I actually want from the finished thing.
The systems are dramatically better than they were. Longer scenes, better movement, increasingly cinematic output, an enormously improved workflow. And for my federation it has still never become economically or operationally important.
Not every capability matters to every person
This is worth emphasising, because these entries are not an argument that every form of generative AI becomes equally important to everyone.
They do not. This ledger is the history of one particular federation.
Images became extremely important. Writing became extremely important. Software, businesses, interfaces, calls, analyses and actions became important.
Video never really did. I have not generated meaningful revenue from it. I have not built a business around it. I have never found myself depending on it in daily productive life.
I still use it. I still experiment with it. Twelve thousand artifacts exist because of that experimentation, and they are useful mainly as evidence of the progression.
The pattern this entry confirms
My preferred form of video generation is still the simplest one: hand an AI a completed artifact — an article, usually — and let it determine how to turn that source into a video.
Which is the recurring shape of this entire story.
The more a technology requires me to become the video editor, the cinematographer, the animator, the prompt engineer or the production technician, the less interesting it is to me.
The more it can simply understand the objective and generate the artifact, the more interesting it becomes.
Video is improving in that direction. Compared with everything else in this federation, it is still a relatively minor branch.