Stand-In is a lightweight identity-control adapter for Wan2.1‑T2V‑14B. It trains only about 1% additional parameters (153M in v1.0), significantly improves identity consistency while preserving naturalness, and can be stacked with community LoRAs.
It is a lightweight, plug-and-play identity control module rather than a traditional style LoRA, but it can be loaded alongside community LoRAs. Target use cases include identity-preserving text-to-video, and it also supports subject-driven generation, pose-guided generation, stylization, and face swapping.