Filmmaker shares full workflow for a 1:42 AI dialogue scene: master voice track + timestamps
DavidmComfort · x · 2026-11-08
Creator David Comfort shares a detailed postmortem of producing a 1:42 AI-generated dialogue scene:
What he did
- Rebuilt the scene as a simpler basement set (orange-moon window, oil lamp, frosted empty tank);
- Generated one master ElevenLabs dialogue track for all 24 lines;
- Rescripted using the reference clip's techniques and cut it into four generations of 26-28s;
- Each generation used a black video carrying the dialogue, a timed transcript, character sheets, a room sheet, and a timestamped shot list;
- Assembled the 1:42 cut with steady room tone, three quiet foley hits, no music.
What worked
- One master track kept voices consistent: pitch within 3 Hz on nearly every line;
- Timestamps for both lines and shots pulled timing within 0.5s;
- Black video + transcript worked on G1 and G3;
- Prompt fixes for duplicate Zoe, plain skin, empty tank all landed.
What didn't: the first segment (truncated). Takeaway: extended AI dialogue is hard; consistency comes from a master audio track plus timestamp constraints.
More from coding & agent
- Mistral Studio ships big upgrade: one-command deploys, traces turned into evals — testingcatalog · 2026-11-08
- Letting an agent wander for 45 minutes: Brownian motion with sudden bounds of Objective — generativist · 2026-11-08
- One skill file replaces 8 MCP servers: researching 50 sites for under $5 — Left_Grapefruit5205 · 2026-11-08
- Big-MoE-Small-Box: serve 361B DeepSeek and 321B GLM from one 24GB GPU via NVMe experts — casper_hansen_ · 2026-11-08
- Karpathy: forcing agents to work is AI's biggest mistake — but critics say that's 18 months stale — jasonkneen · 2026-11-08
- Fireworks launches Nexus: routes coding tasks across open and closed models, cuts spend 50% — sophiamyang · 2026-11-08