1. Workflow Overview

This "Mecha Transformation" workflow uses WAN2.1 video model to dynamically convert portraits into armored warrior videos. Key features:

• Background preservation

• Smooth human-to-mecha morphing

• 720P output (16FPS)

2. Core Models

• Main Model: FP8-optimized Wan2_1-I2V-14B

• LoRA: WAN2.1 Mecha Transform (specialized)

• Video VAE: FP32 precision decoder

3. Key Nodes

Dependencies:

• Text encoder umt5-xxl-enc-bf16

• CLIP vision model open-clip-xlm-roberta

4. Pipeline Stages

Stage 1: Initialization

• Load WAN2.1 model trio

• Inject mecha LoRA at 1.0 strength

Stage 2: Motion Control

• Positive prompt: "woman wears mecha suit"

• Negative prompt filters artifacts

Stage 3: Rendering

• dpm++ sampler (20 steps)

• Temporal consistency with TeaCache

5. I/O Specification

• Inputs:

  • Portrait image (e.g., 1024x1440 PNG)

  • Fixed seed 884841285240243

• Output:

  • 720P video with CRF19 compression

6. Critical Notes

⚠️ Requirements:

• 16GB+ VRAM (FP8 optimized)

• CUDA 12.1+ recommended

🔧 Pro Tips:

• Adjust SLG scale (0.1-1.0) for morphing intensity

• Enable sageattn in model loader for speed boost