What's this?

My previous Lock-On Stabilization workflow (Lock-On Stabilization using Florence2 and SAM2) Acropped the video to match the subject's shape. This workflow takes the opposite approach: instead of cropping, what if we use **Wan2.1 VACE outpainting** to fill in the areas that become visible when the camera tracks the subject?

Workflow

This workflow consists of two main steps: Lock-On Stabilization and Wan2.1 VACE Outpainting. While they're connected in this workflow, users with 12GB VRAM or less should run these steps separately.

Lock-On Stabilization using Florence-2 and SAM2

• 🟥 Resize images  
• 🟪 Add green padding around all edges  
• → This green area will later be converted to masks  
• 🟩 Object detection with Florence-2 + segmentation with SAM2  
• 🟦 Extract mask position and size, then crop so the mask stays centered  
• 🟨 Receive X(Y) axis coordinates as lists and smooth them using Kalman filter  

Wan2.1 VACE Outpainting

• 🟦 🔧 Mask From Color node converts green areas to masks  
• 🟦 Convert green areas to gray using the masks (VACE requires gray areas for outpainting)  
• → Then standard Wan2.1 VACE processing