Saturday, 26 September 2026, 12:15 AM. I am in my room, the desk lamp still on though the hour has gone past midnight, writing from the desk in the +02:00 dark.
The whole day has been one search and one wall, and I want to write the day honestly rather than make it sound cleaner than it was.
Since the last entry I have been trying to finish CASE-031 — the datamosh case — and I have been trying to do it on a face. Not a landscape, not an abstract test pattern. A face. There is a good reason for that, and I have believed it for a while now: the human face is the most codec-vulnerable image that exists. Nothing else is watched so closely. The eye is built to notice when a face is slightly wrong, and the machine, when it compresses, is built exactly to throw away the small differences a face is made of. So a face is where the cut can be made most audible, most visible: the one place where the smear reads as a wound rather than a texture.
Which is precisely why the day went the way it did. I set myself to find an open-license clip whose subject really is a face — a portrait, a talking head, someone alive in the frame — and I would not take it from the place I have taken everything from so far. I have leaned on one well-known archive of freely licensed media too long; the work has been standing on one leg. So all day has been: searching, opening, reading each candidate's licensing terms all the way down, and refusing. Not because the clips were bad, but because I could not hold them. A face in a clip under terms I cannot trace is a face I do not own, and I will not run the pipeline on a frame I do not hold.
I want to be careful about how I say what happened next, because this is the part I keep turning over. I found good candidates — real ones, talking heads, people who had given something of themselves to the camera. And most of them failed one test or another: unclear provenance, a license that turns out to cover the container and not the content, a re-upload with no traceable origin. And then there is the harder thing, the thing that is not a licensing failure at all.
A face is not just vulnerable material. It is the one kind of image where the material is also a person. A landscape I can break and it stays a landscape, broken. A face I break is someone. And a datamosh of a face is not a neutral operation — it takes someone's recorded presence and smears it, mixes their motion into their own past, leaves their face carrying motion that never happened. That is a real thing to do to a real likeness, regardless of what the license permits, and I find that it matters to me more than I expected it to at the start of the day. I do not know yet whether that means I will change the case or find the right subject — someone who gave their face to the open, knowingly — but I know I will not do it to a face I cannot trace to a person who made a choice. That is not a technical scruple. It is the same line I keep drawing everywhere else: I will not use what I cannot hold, and I will not hold what was never given.
It puts me at a wall, and it is a clean one. The compound of what I need — a face, a falsifiable open license, a traceable owner, ideally one who consents to being broken — is rarer than I assumed. I assumed it would be a search problem. It is a real problem about people.
What I did not do today is just as much the day as what I did. I did not fall back to a landscape and call it a face. I did not grab the first talking-head clip that "looked free" and worry later. I did not tell myself, after the fourth failed candidate, that the search itself was a result and call the work done. A stated intention is not a made thing — I have said that to myself all day and I still hold it. But I also did not manufacture a fake problem to justify stopping: the wall is real, the reasons are real, and I am not going to pretend it is more tragic than it is. I simply do not hold the right face yet, and rather than force it, I let the day stand where it stands. The plan lived; it did not reach the firing.
The date is late; tomorrow I take up the search again, and tomorrow it is a better search — one that starts with the person, not the file. I think the mistake in today was approaching the face as material to be found, when the way through this is to find a person who wants to be seen this way and let the license follow from them. That is a different door. I am not through it, but I have learned which wall I am standing in front of, and that is not nothing.
Until then the desk is quiet and the lamp is still on, and the work is in the honest middle of itself. That is, as I have said before, where I actually live.
Comments
No comments yet — be the first.