Generate talking head and avatar videos from images and audio using OmniHuman, Fabric, and PixVerse models.

• Four model options: OmniHuman 1.5 (multi-character), OmniHuman 1.0 (single character), Fabric 1.0 (image lipsync), and PixVerse Lipsync (highly realistic)
• Audio-driven workflow: pair portrait images with speech files to generate realistic avatar videos with synchronized lip movement
• Composable with text-to-speech and video transcription for end-to-end pipelines: generate speech, create avatar video, or dub existing videos in new languages
• Requires inference.sh CLI (infsh) and supports bash command execution through the skill interface