Discovery on ElevenLabs v4

Four complete soundtracks per ad, one per tag. Pick a take, nudge the timing, set the loudness, and the page tells you whether it costs a mux or a re-render.

Bury St EdmundsQA gateBuilt tonight on the v4 base end to end, all four formats rendered. Approve the After Effects mix, or pick a different one and apply it with ffmpeg, no re-render. Winchesterbuilt tonightMade from scratch: data, 12 v4 takes, four soundtracks, After Effects picture, all four formats. Plays the real ad against whichever mix you choose. Winchester, v4 basenewThe same reads laid on 000000-Lucas v4, the common narration regenerated in v4. Gates from your markers, reads matched to the v4 base at -19.2 LUFS. A/B this against the one above. Gerrards CrossThe same four-angle build, audio only. The comparison test. Leamington SpaYour locked recipe applied cold: 3 generations, no alternates.
After Effects owns the picture. The mix it rendered with is the deliverable, and ffmpeg only lays a different soundtrack on when you reject that mix, by video stream copy, about 2 seconds per format. A caption word moving more than 0.30 s is the only audio-side change that sends it back to After Effects.