AI dialogue short post-mortem: Seedance prompt leaks, ElevenLabs pause limits and failed voice patching
DavidmComfort · x · 2026-11-05
DavidmComfort's what-worked/what-didn't list for his AI conversation short.
What worked: three reference sheets per clip, one beat spec driving both voice and video prompts, transcript-checked short ElevenLabs tags re-spaced in post, a staging clause holding the 180 rule, cutting to the other person at every seam, steady music bed with sparse foley.
What didn't: patching voices afterward with speech-to-speech or TTS fillers; Seedance narrating facial-expression words during silent audio ("A faint smile softens"); matched first frames leaving a visible pop; ElevenLabs pause tags buying only 1.7s and long directions breaking takes. He urges people to start using agents for this workflow.
More from Multimodal
- Dev uses new image model to redo game art, says OpenAI has no image-gen rival — CtrlAltDwayne · 2026-11-05
- VA-TDM: Velocity Adversarial Distillation Sets SOTA for Few-Step Diffusion Models — Yihong Luo · 2026-11-05
- AI-generated anime music video 'Jealousy' showcases cinematic AI animation — Few_Beautiful2589 · 2026-11-05
- 31.7MB Open-Weights TTS Model Woof Powers an Entire Browser Game, Fully On-Device via WebGPU — brandon_xyzw · 2026-11-05
- CapCut AI fills an unfilmable ad shot with a reference-frame workflow — JaynitMakwana · 2026-11-05
- Veo-generated simulated flight over Mars, built from NASA imagery with original score — DarkMatterProd_ · 2026-11-05