Wan 2.2 Skin Morph LoRA Guide
How to use the SkinMorph LoRA for Wan 2.2 I2V 14B to create smooth identity-transformation and morphing video effects in ComfyUI.
On this page
Wan 2.2 Skin Morph LoRA Guide
SkinMorph is a LoRA built for Wan-AI Wan2.2-I2V-A14B that turns ordinary image-to-video prompts into morphing identity-transformation sequences. Instead of a simple face swap, it animates a gradual skin-level transition: texture, tone, and facial structure blend across frames to turn one person into another, a different gender, or a stylized character.
- Generates smooth skin-morphing video transitions between identities.
- Trained on Wan2.2 I2V 14B with separate low-noise and high-noise LoRA weights.
- Uses a single trigger phrase: Skin morph.
- Best suited for character-change, transformation, and special-effects content.
Sample output: upper-body morph sequence with mask-like skin texture transition.
What the LoRA does
In sampled outputs, the model preserves the subject’s pose, clothing, and lighting while animating the face and exposed skin through a textured, mask-like metamorphosis. One video shows a young man with curly hair transforming into a figure with a dark blue, mask-textured face; another shows a suited man whose face shifts to a light gray scaled texture; a third shows a woman in traditional dress with a dark floral-patterned face morph. The motion remains coherent across the transition rather than cutting between two static identities.
How it was trained
The LoRA targets the Wan-AI Wan2.2-I2V-A14B base model and is delivered as two safetensors files: a low-noise variant and a high-noise variant. That split follows common Wan I2V LoRA practice: one adapter handles earlier denoising, the other refines later stages, which helps keep morphs smooth instead of noisy or flickering. The model card lists CC0-1.0 as the license.
Recommended settings
- Trigger: include
Skin morphin the prompt. - LoRA strength: start around 0.8–1.0.
- Workflow: ComfyUI Load LoRA node after the base Wan I2V loader, feeding both low-noise and high-noise files.
- Prompt style: describe the target identity or character directly; the LoRA handles the morph rather than requiring elaborate transition wording.
Example prompts from the author:
Man morph to a basketball player. Skin morphMan morph to a woman. Skin morph.Boy morph in a james bond suit. Skin morph.
ComfyUI setup
- Download both
[WAN2.2]SkinMorph_Redmond_low_noise.safetensorsand[WAN2.2]SkinMorph_Redmond_high_noise.safetensors. - Place them in
ComfyUI/models/loras/. - Load your base Wan 2.2 I2V model.
- Add a Load LoRA node and connect both LoRA files.
- Attach the LoRA model output to your KSampler and adjust
strength_modelbetween 0.8 and 1.0.
Common errors and fixes
Error | Cause | Fix | Source
|---|---|---|---
Flickering / unstable morph | LoRA strength too high or single-file use | Start at 0.8; use both low-noise and high-noise files | Model card Weak transformation effect | Trigger phrase missing | Add Skin morph to the prompt | Model card Poor identity blending | Base model mismatch | Ensure base is Wan2.2-I2V-A14B | HF README
FAQ
Q: Does this work in Automatic1111? A: The published instructions target ComfyUI. Wan I2V workflows are primarily built around ComfyUI custom nodes, so expect to route through a ComfyUI setup rather than the standard A1111 txt2img pipeline.
Q: Can I morph between specific photos? A: The LoRA is trained for prompt-driven identity transitions. Best results come from descriptive prompts naming the target identity or character, not from direct image-to-image reference pairs.
Q: Is the model free to use? A: Yes. It is released under CC0-1.0.