Datamosh transitions, on a phone

That is why datamoshing looks the way it does. It is not noise or a filter; it is a picture being dragged around by motion that belongs to something else. It smears in the direction the new shot is moving, it holds perfectly still when the new shot is still, and it is made of the exact pixels of the shot that just ended.

Why that is different from a glitch overlay

Most “glitch effects” in mobile editors are artwork: coloured bars, RGB splits and scan lines composited on top of your video. They look the same on every clip, because they have nothing to do with your clip. You could screenshot the timeline and predict the result.

A datamosh cannot be predicted that way, because it is computed from your two specific shots. Cut from a locked-off tripod shot into a handheld whip and the picture is torn off the screen; cut into a static shot and it barely moves at all. Same effect, same settings, completely different result. That is the honest test of whether an app is datamoshing or decorating.

The Glitch transition is a transition between two adjacent clips on one track, not an effect you apply to a clip. Two clips in a row is the entire prerequisite: nothing is stacked on a layer above, nothing overlaps, no audio is detached, and neither clip gets shorter.

What each clip contributes

The outgoing clip gives one frame. The app reads the frame you were actually looking at when the cut arrived, freezes it, and uses it as the reference picture. That still is what the incoming clip's motion data gets applied to, which is why the smear is made of the shot that just ended.

The incoming clip gives the movement. Its picture-carrying frames are dropped and only its motion frames are kept. Those describe how things moved in that shot, and they are now moving pieces of the previous shot around.

Why it ends the way it does

Alongside movement, each frame carries a small correction to the picture. Those corrections accumulate, so the borrowed image is repaired a little at a time and the incoming clip gradually reassembles. That decay is the ending, and deliberately the only one: the mosh runs the entire length of the incoming clip, so it never resolves with a cut from a fully smeared frame to a clean one.

The strength slider

Strength decides how coarse the compression is, and therefore how much correction gets through. Turn it down and the picture heals sooner; turn it up and the mosh takes longer to resolve. It is the same material either way, just slower; strength is not a second effect layered on.

The middle of the slider is the neutral setting, so a Glitch transition applied and left alone is a normal datamosh rather than an extreme one. In a typed instruction, a size word is what moves it: a subtle glitch, a heavy datamosh.

The duration slider crossfades the audio

It is labelled Audio crossfade, and that is all it does. A hard audio cut underneath a picture that is smearing sounds wrong, so the linked audio is blended across the join. The picture's duration is not affected and neither clip is trimmed.

The effect is computed from your footage, so the footage is most of the work:

• Cut into movement. The mosh only travels as far as the incoming shot moves. A whip pan, a fast walk or a camera turn will haul the previous frame right off the screen; a locked tripod shot will barely disturb it.
• Hold, then go. Cutting into a shot that is still for a second and then moves gives you the best version of this: the previous frame sits there looking almost untouched, and then is torn apart. That pause is the effect working, not a delay.
• Give it room. The mosh runs the length of the incoming clip and decays as it goes, so a very short incoming clip cuts away before the decay reads.
• Use Bloom sparingly. It never resolves, which makes it an ending rather than a join. Two Blooms in a row is usually one too many.
• Contrast the shots. The smear is the outgoing frame, so a cut between two visually different shots reads far more clearly than a cut between two similar ones.