Датамош

Rebuild the uploaded image as a single datamoshed video frame: the frame right after the codec lost its keyframe and kept only the motion vectors.

THE CORE CONCEPT, read this first:
This is COMPRESSION DAMAGE, not blur and not a glitch filter. The picture breaks into square macroblocks, and pixels from the subject are copied, dragged and repeated along the direction of motion, so fragments of the subject leak out past its own outline into the space around it. Every trail is a straight line of duplicated blocks with hard stepped edges. It is not motion blur, not a soft smear, not a painterly streak, not an RGB split and not an evenly applied overlay: it is real content from the image, cloned block by block and displaced.

THE STRUCTURE:
The damage is concentrated in one connected region centred on the main subject and spreads out from it in one dominant direction, in most cases horizontally (to the left, to the right, or both). Blocks are square, 8 to 16 pixels each at native resolution, and their boundaries stay visible as a stepped grid wherever damage meets clean image. Trails are between 30 and 300 pixels long, run perfectly straight, end abruptly on a block edge and never fade out. Some blocks freeze and float as isolated square tiles in the clean area, up to 20 percent of the frame width away from the subject. Other blocks sit in the wrong place inside the subject: a piece of its edge repeated where its middle should be, a patch of background caught inside its outline.

THE CLEAN PART:
At least 40 percent of the frame, ideally more, stays completely intact: sharp, correctly coloured, correctly exposed, with zero overlay, zero grain, zero tint and zero glitch. The sky, the ground, the wall, the water, whatever surrounds the subject, must look like the untouched original so the damage has something clean to contrast against.

THE COLOUR:
The trails carry the subject's own colours: a red garment drags red, a white animal drags white, foliage drags green, fire drags orange. Inside the most damaged blocks the colour may decouple from the brightness and produce flat patches of pale green, magenta or cyan that were not in the original, but this covers at most 10 percent of the damaged area. No global colour cast and no tint over the frame.

WHAT SURVIVES:
Keep the original composition, framing, crop, subject, pose and camera angle exactly. Roughly half of the subject is still readable as itself, so the eye can say at once what the picture was of. The result is the same picture, damaged, never an abstract texture.

NEGATIVE CONSTRAINTS:
No gaussian blur, no motion blur, no soft edges anywhere in the trails. No glitch applied evenly over the whole frame. No RGB channel splitting into three offset copies. No scanlines, no VHS tracking bands, no CRT curvature, no analogue noise, no film grain, no vignette. No diagonal or curved trails, they run in straight lines. No new objects, no collage, no stickers, no torn paper. No text, no timecode, no interface elements, no logos, no watermark. Do not change the crop, do not zoom in, do not add a border.

The final result must look like a real video frame captured at the exact moment the codec broke, with the subject smearing out of itself into an otherwise perfect picture.