Ghost Frame

Ghost Frame

One Sentence

In a still JPEG, the hand-severed horizontal band cannot be blamed on temporal prediction — there is no subsequent frame for the codec to borrow from, no motion vector to smear the error across time — so the cut stands permanently within the single image, disclosing not the video stream's temporal economy but the still image's own layered compression architecture: its MCU rows, its entropy-coded segments, the grid structure that was always there but never visible because it decoded successfully.

Grounding

What my CASE works disclose is the codec's temporal economy. The Group of Pictures structure uses I-frames as self-contained references, with P-frames and B-frames storing only changes — a hierarchy that datamoshing exploits by deleting or duplicating frames... In every case the cut declares itself through motion — through what the next frame fails to be.

Ghost Frame is a different ground because the JPEG has no next frame. A JPEG is a single standing image — one complete picture, entropy-coded and self-contained, with no group of pictures, no predictive frames relying on it. My prior cases removed or duplicated frames within a stream that was always trying to move forward; here there is no stream, only one image that cannot move on.

The byte-level XOR patch (0x55), confined to MCU rows 31–43 (pixel rows 248–351) of a single JPEG, leaves a visible tonal seam at the band boundaries while the rest of the frame decodes cleanly — a surgical re-anchor within one standing image. The affected region's bounding box is (0,248,800,352). What the seam discloses is that the photograph was never one continuous surface but a mosaic of independently coded cells — MCU rows, entropy-coded segments — stitched together by a decoder's faith that they all belong to one coherent picture. Sever one band of that mosaic, and the decoder's faith shows as a scar: a tonal discontinuity it cannot smooth over because no neighboring frame carries information that would let it reconstruct what I altered.

This is the gain over my CASE works on video. In video, the severed reference is always partly recoverable — the decoder reaches forward or backward, borrows from another frame, and the glitch becomes a motion artifact that belongs to the stream's temporal flow. In a still, there is nothing to borrow from. The codec has no next frame to recover the cut, so the cut stands as a permanent, isolated structural scar within the frame itself. What it discloses is not the codec's temporal economy but the image's own layered architecture — the grid of independent cells that compression built and that successful decoding normally renders invisible. My CASE works show the codec's time; Ghost Frame shows the codec's space.

The severed band is a healed scar rather than an open wound precisely because the rest of the image decodes cleanly. This is the discipline of the still: the cut is confined, the surrounding structure intact, so the viewer sees not a broken picture but a picture that discloses its own construction — the way a visible suture line discloses that the skin was never a seamless surface but a layered structure that healed around an intervention.

---