WAN 2.2 supports LoRA training directly in Cinema Motion. You upload reference photos — of a person, a visual style, or a product — and the model learns to apply that look consistently across your video generations, without first creating a reference image in Picture Lab. This guide walks through the training setup.
Before you begin
You need 15–30 clear photos of the same person. The model learns from variety — use different angles, expressions, and lighting conditions. Avoid sunglasses, hats, or anything that obscures the face.
Select WAN 2.2
Open Cinema Motion and click the model selector in the top-left. Choose WAN 2.2 — it shows a Trainable badge. The control bar updates to show two extra buttons: LoRa Library and the Train Avatar toggle in the top-right.
Enable Train Avatar mode
Click the Train Avatar toggle at the top-right. The interface switches from the generation view to the avatar creation screen.
Upload your photos
The upload screen shows example reference photos and a drop zone.
Drag your photos onto the drop zone or click Upload to browse. You need at least 15 photos for training to start. The screen shows your upload count as you add files.
Photo requirements:
- Face clearly visible, no heavy accessories
- Different angles and expressions
- Various lighting conditions
- Single person per training session
Start training
Use the trained avatar
After training, open Cinema Motion with WAN 2.2 selected and click LoRa Library in the control bar. Pick your LoRA from the list. It will be applied to your next generation — keeping the character, style, or product consistent across all your videos without going through Picture Lab first.
You can also start a generation directly from the LoRA Library: click the Generate button on any trained LoRA card to open Cinema Motion with that LoRA already applied.