One photo.
Two finished directions.

Variation: selective regeneration

A simulated client brief asked for the product in shot 2 swapped for a fictional Verde variant and the spoken line in shot 3 redone, with shots 1 and 4 left alone. Shots 2 and 3 were regenerated as new takes from the same setup, references and timing; shots 1 and 4 use the identical source files, verified by hash.

This demonstrates selective regeneration with a new take. It does not yet prove the seamless in-place replacement the brief asks for: the motion inside the two regenerated shots differs from the original takes. Measured on the finished exports: shots 1 and 4 are cut from byte-identical source files (hash-verified). In the final MP4, 137 of shot 1's 180 frames and 89 of shot 4's 90 frames are hash-identical to the original export; the remaining frames sit on the cuts (shot 1's at 48.9 dB or better, shot 4's one frame on the 17 second cut). The audio mix is re-normalised as a whole, so the untouched shots' audio differs only by overall gain. Shots 2 and 3 differ throughout, as regenerated takes should.

The Variation tab plays the original on the left and the variation on the right, in sync, then a three-second summary. The measurements are in the change log.