ComfyUI WAN 2.2 - S2V

ComfyUI with WAN 2.2 (s2v) on NVIDIA GPUs via RunPod.
Includes optional model installs (s2v & i2v), upscalers, interpolation, LoRA support, and helper utilities.

Version: v0.0.2

---

Features

• Base: nvidia/cuda:12.8
• WAN 2.2 Support:
  • i2v 14b (image → video)
  • s2v 14b (sound → video)
  • Enabled via environment variables
• Models / Checkpoints:
  • umt5_xxl_fp8_e4m3fn_scaled.safetensors for CLIP
  • wan2.2_vae (default) or wan2.1_vae (optional)
  • 4xlsdir.pth upscaler
• ComfyUI Built-ins:
  • LoRA support (disabled by default)
  • Downloader for CivitAI models
    • Requires CivitAI token + version # via environment variables
• Interpolation: RIFE frame interpolation
  • Default: 16 → 30fps (for social media)
  • Configurable to arbitrary FPS
• Sampler Options: ClownSharkSampler for high/low noise (i2v workflows)
• RunPod Optimized: Pre-configured for pod templates and GPU persistence

---

Getting Started

Prerequisites

• RunPod account with GPU access

GPU Selection

• When Deploying a Pod, expand Additional Filters and select CUDA Version 12.8
• Minimum recommended GPU spec is 48 GB VRAM (e.g. L40S)

Environment Variables

Control optional installs and features using env vars:

Variable	Accepted Values	Description
INSTALL_WAN22_I2V	true / false	Install WAN 2.2 i2v 14b
INSTALL_WAN22_S2V	true / false	Install WAN 2.2 s2v 14b
CIVITAI_TOKEN	<token>	API token for CivitAI downloads
LORAS_IDS_TO_DOWNLOAD	<version_id>	Specific version # to fetch

---

Usage

RunPod

1. Launch a pod with CUDA 12.8 base.
2. Set environment variables (see above).
3. Ensure ComfyUI & JupyterLab service ports exposed (default 8188 & 8888).