Wan 2.2 + ComfyUI on 8GB VRAM: Fast Setup & Image-to-Video Test

🚀 Wan 2.2 is a newly released image-to-video AI model that can generate cinematic video from a single still image. The best part? You can run it in ComfyUI with just 8GB of VRAM.

Here’s a quick setup overview based on my full video tutorial.

---

🔧 What You Need:

• Wan 2.2 GGUF Model (5B for 8GB VRAM)

• LoRA: Lightx2v for motion behavior

• VAE: AutoVAE KL-f8

• Text Encoder: UMT5

• Workflow JSON from Hugging Face

📁 Place the files into these ComfyUI folders:

bash Copy codeComfyUI/models/unet        → GGUF model  
ComfyUI/models/lora        → LoRA  
ComfyUI/models/vae         → VAE  
ComfyUI/models/text_encoders        → Text encoder

---

🧠 Running the Workflow:

1. Import the Hugging Face .json workflow

2. Load your image (I used a wolf)

3. Set your prompt (e.g. “howling wolf in a storm”)

4. Hit execute — AI-generated video from your image

---

🛠️ Fixing the sageattention Error:

Got this error?

codeModuleNotFoundError: No module named 'sageattention'

I used Gemini CLI to fix it automatically — just paste the error, and it’ll install the missing module for you.

---

🎥 Watch the Full Video Tutorial:

See the full walkthrough, node setup, and final output here:
👉 https://youtu.be/7hUO6KhUsvQ

---

This workflow makes AI video generation accessible without high-end hardware. If you're experimenting with ComfyUI or open-source video models, give this a try — it works better than you'd expect on mid-range machines.