One pass now does the people, the product and the speech
Seedance speaks the ported script. Given the three lines with their seconds, it says them in the man’s new voice, and it re-lettered the donor’s burned-in caption to match. That removes the silence, the mute stage and the revoice stage in one move. The bottle is also correct now: white with the blue trigger top, the INSTANT SLIP+SHINE label and the yellow starburst, held steady across the clip. Describing the bottle to match the photo I was already attaching was the whole fix.
Two things to fix before this ships. The brand name comes out as "EB" rather than "Ebb", both spoken and in the caption it draws, which the method already anticipates: simplify or respell any word the model mispronounces. And the small print on the label is approximate, so a shot that needs a pin-sharp label wants our own packshot composited instead.
Their original, first 6 s
The celebrity stylist, his real voice, the teal bottle.
Ours: his real voice laid back on
New man, new face on her, her hair untouched, and the correct bottle held steady. The audio is the donor’s own track put back over the new picture, so every word is exactly as he said it.
Ours: Seedance speaking our script
No donor audio at all. Seedance generated the voice from the three ported lines. Whisper reads back line 1 word for word and line 3 as far as the clip runs. It also redrew the burned-in caption with our words.