wan-scail-replacement

Guides SCAIL-2 in-video character replacement on WAN 2.1 with framing, tuning, and compositing rules.

715|111|Updated Feb 15, 2026
One-click install
npx skills add https://github.com/artokun/comfyui-mcp --skill wan-scail-replacement
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: wan-scail-replacement
Source: https://github.com/artokun/comfyui-mcp/tree/main/plugin/skills/wan-scail-replacement
Command: npx skills add https://github.com/artokun/comfyui-mcp --skill wan-scail-replacement

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

SCAIL-2 character replacement renders in ComfyUI fail in non-obvious ways—oversized characters, leaked driving-subject details, and broken compositing—that each cost a full multi-minute render to discover. This Skill documents the measured pitfalls so you avoid re-renders.

Core Features & Use Cases

  • Reference Framing Rule: Explains that in replacement mode the reference image's framing controls the output character's scale, with measured pixel data showing how to pad/reframe references to match the driving subject's size.
  • Tuning Trade-off Guidance: Documents how distill LoRA strength and ModelSamplingSD3 shift settings trade color/detail quality against driving-subject detail leakage onto the replacement character.
  • Compositing Anti-Pattern: Explains why post-hoc mask compositing to fix color is a regression, since SCAIL-2 resolves color, edges, grip, and occlusion jointly.
  • Use Case: You run the official video_wan21_scail2_character_replacement_int8 template and the swapped character renders 1.7x too large; this Skill tells you to reframe the reference image so the person occupies the same fraction of frame height as the driving subject.

Quick Start

Use the wan-scail-replacement skill to diagnose why my SCAIL-2 character replacement render has the wrong character scale and how to fix the reference image framing.

Frequently Asked Questions about wan-scail-replacement

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I fix wrong character scale in SCAIL-2 character replacement?

In replacement mode, the reference image's framing controls the output character's size. Pad or reframe the reference onto a canvas so the person occupies roughly the same fraction of frame height as the subject in the driving video before rendering.

What is SCAIL-2 character replacement in ComfyUI?

SCAIL-2 is a WAN 2.1 14B-based model that replaces the person in a driving video with a character from a reference image, end-to-end with no pose maps. The official template uses WanSCAILToVideo with SCAIL2ColoredMask and a SAM3 mask.

Why does the original subject's details appear on my replacement character?

Running the lightx2v distill LoRA at 0.8 with ModelSamplingSD3 shift 5 increases adherence to the driving video, leaking original-subject details onto the replacement. Raise the LoRA strength and shift back toward 1.0 and shift 8.

Should I composite SCAIL-2 output over the original video to fix color?

No. Mask compositing produces halo artifacts and severs grip relationships the model solved coherently. SCAIL-2 resolves color, edges, grip, and occlusion jointly, so correcting one in isolation breaks the others.

How much VRAM does SCAIL-2 character replacement need?

The int8 template build peaks at about 22.2 GB VRAM at 576x1024 with 81-frame chunks, fitting on a 24 GB card. It uses the SCAIL-2 int8 model, DPO LoRA, distill LoRA, WAN 2.1 VAE, UMT5 text encoder, and SAM3.