Wan2.2 I2V · Two-Stage Sampling · RIFE · 4x-UltraSharp · 1664×2432 · 60fps

## Workflow Overview

This workflow converts a character image into a high-resolution 60fps video using Wan 2.2 I2V with a full post-processing pipeline.

Input:

Output: 1664×2432 @ 60fps video

---

## Pipeline Stages

| Stage | Node | Detail |

|-------|------|--------|

| 1 | WanImageToVideo | I2V with CLIP Vision + text prompt guidance |

| 2 | KSampler Stage 1 | High noise (8 steps, Euler, denoise 0.5) — establishes motion skeleton |

| 3 | LatentNormalize | Aligns Stage1→Stage2 latent distribution |

| 4 | KSampler Stage 2 | Low noise (12 steps, Euler, denoise 1.0) — refines details |

| 5 | VAEDecodeTiled | Tile-based VAE decode (tile=1024, overlap=128) |

| 6 | RIFE VFI | 2× frame interpolation → smooth 60fps motion |

| 7 | ImageUpscaleWithModel | 4×-UltraSharp upscaling to 1664×2432 |

| 8 | Image Lucy Sharpen | Final sharpening pass |

---

## Key Settings

- Unet (GGUF): wan22EnhancedNSFWV2Q8H + Q8L (FastMOVEv2)

- VAE: wan_2.1_vae

- Clip: umt5_xxl_fp8_e4m3fn_scaled

- RIFE: rife49.pth, fast_mode + ensemble, 2× multiplier

- Video output: HEVC NVENC, 60fps, pingpong=yuv420p

---

## Tips

- Two-stage sampling dramatically improves motion coherence vs single-stage

- VAEDecodeTiled is essential for resolutions above 1024 to avoid OOM

- RIFE 2× is the sweet spot — 3× introduces visible ghosting

- Lucy Sharpen at default adds crispness without artifacts

- Shift=5 on ModelSamplingSD3 matches Wan 2.2's native motion profile