SadTalker

Nossa avaliação — SadTalker para minidramas

SadTalker — a CVPR 2023 project from Xi'an Jiaotong University and Tencent AI Lab — turns one portrait photo plus an audio track into a talking head, using 3DMM motion modeling with an optional GFPGAN face enhancer. For short drama our honest read: it's the classic budget primitive for narrator inserts and stylized flashback faces, not scenes carrying emotional weight.

Against LivePortrait, SadTalker is audio-driven where LivePortrait is performance-driven — no actor required, just a voice file, which suits fully synthetic pipelines. It's Apache 2.0 and self-hosts fully offline, a real privacy plus for unreleased scripts. But major updates date to 2023, so we'd treat it as stable-but-aging; newer hosted lip-sync like Sync will out-render it.

• Photo plus audio to talking head
• 3DMM motion, GFPGAN enhancement
• Apache 2.0, fully offline
• Stable Diffusion WebUI extension

Escrito pela equipe editorial da DramaSo Tools, em 2026-07 — uma leitura editorial do site oficial, não uma nota testada na prática.