Treasury myths — explainer reel
Video Audio — voiceover
Script from post — hook “£1.2M sat idle last year” → 4 beats → CTA outro · 40s target
Edit in planner
Clips 3 · 40s
Set
Seedance 2.0 · 15s cap
Style: kit + reel twist
AI re-plan
Saved
Audio — voiceover inherits the production plan
Voiceover · inherited
0:14
Where the VO lands
Embed routing is automatic. Seedance 2.0 accepts an audio role, so the VO rides INTO the generation — the model lip-syncs pacing to it and the take comes back with sound (the “embeds into clip” badge). On a model without the audio role the badge flips to “overlay at assembly” and the VO is layered when the final video is assembled.
Takes — Clip 2
2 takes
0:15
take 1 · run 1silent
0:15
new
take 2 · run 2 VO 2m ago
Audio is one section, not a stage. VO text seeds from the clip’s narration beat, the brand voice (Maya) rides in from the production plan, and Generate VO is a single ElevenLabs call. Takes with embedded audio carry the VO marker — take 1 predates the VO and stayed silent.
Audio states — the full per-clip ladder
1 · Idle — seeded from narration
2 · Generating
3 · Ready — overlay variant
Voiceover
0:14
4 · ElevenLabs not connected
Voice generation needs an ElevenLabs key on this workspace. Connect it in Settings → Connections — or upload a recorded VO instead.
Settings → Connections — the ElevenLabs card (new)
11
ElevenLabs Voice · BYOK
Workspace-level key — every brand in this workspace uses it for voiceover. Each brand picks its own voice.
Stored encrypted. Used only by voice-generation jobs.
This card never existed. The backend resolver (resolveElevenLabsConnection) shipped with the audio subsystem, but the only way to enter a key was the Payload admin. The floor links here from the not-connected state. Brand voice (Maya) is picked on the Brand settings page and rides into every production’s audio plan.