Player not loading? Watch on YouTube
This tutorial explains how to create a talking avatar with SadTalker through Pinokio. The speaker installs Pinokio for their operating system, finds SadTalker in its explorer, and installs it along with its dependencies. The local AI workflow uses a source image and an uploaded audio file. The speaker recommends a square image at 512 resolution and short audio chunks for PCs with limited VRAM.
The demonstration runs on a PC with 8 GB of VRAM. The speaker skips reference video controls, recommending them only for a strong GPU with more than 12 GB of VRAM because processing may crash otherwise. Pose style controls head movement; the example uses 16. Expression scale controls mouth opening and facial reactions; the example uses 1, with a warning that values above 1.5 can distort the face.
The tutorial explains how crop, resize and full preprocessing affect framing. The speaker recommends crop for low VRAM, but demonstrates full with still mode enabled. They recommend FaceVid 2 Vid and a batch size of one below 12 GB of VRAM. For face resolution, the speaker recommends smaller outputs on limited hardware.
GFPGAN is an optional face enhancer. The speaker says it sharpens the animated face but adds 20% to 30% to generation time. The tutorial ends with a generated clip. Generation time depends on the PC, input length and settings.
