Draft page for internal preview — source-backed claims and links must be reviewed before publishing.
Wan 2.2: Alibaba's Open-Source Video Model (ComfyUI & Local)
Alibaba’s open-weight video family, with two model sizes and mature ComfyUI support.
Quick answer
Wan 2.2 is Alibaba's open-weight video generation family, published under the Wan-AI organization on HuggingFace.
It is popular in ComfyUI because it offers multiple model sizes and workflows, including text-to-video and image-to-video variants.
The model weights are free to download and self-host, but your real cost is GPU hardware/cloud compute; hosted Wan services may have separate pricing.
01Wan2.2 TI2V 5B vs I2V A14B
Wan gives you a size choice: TI2V-5B for cheap experiments, I2V-A14B when image-guided quality matters more than speed.
| Wan2.2 TI2V-5B | Wan2.2 I2V-A14B | |
|---|---|---|
| Positioning | Smaller text/image-to-video model | Larger image-to-video model |
| Weights | Open · HuggingFace | Open · HuggingFace |
| Model fee | $0 | $0 |
| Best for | Lower compute tests | Higher-quality I2V workflows |
02Generation modes
Wan’s draw is that you can run it yourself and rewire the workflow.
Text to Video
T2VGenerate short clips from prompts with open-weight local workflows.
text → video
Image to Video
I2VAnimate an input image while steering motion with a prompt.
image + prompt → video
ComfyUI workflows
ComfyUICommunity nodes and workflows make it practical for local experimentation.
workflow + weights → video
03Run locally
Wan 2.2 is an open-weight local option. Budget depends on model size, resolution, frame count and quantization; HuggingFace GPU examples start at $0.40/hour.
VRAM by GPU
| Setup | Concrete cost | Best for | Notes |
|---|---|---|---|
| Local GPU | $0 model fee + your hardware | Private/offline generation | Consumer GPUs may need quantized workflows |
| Cloud GPU | HF examples: T4 $0.40/hr, L4 $0.80/hr, A10G $1.00/hr | Occasional heavy jobs | Cheaper than buying hardware for low volume |
| Hosted GPU | HF Spaces CPU free; GPU from $0.40/hr | No setup | Use HuggingFace or another GPU host |
04Official pricing & registration
Wan 2.2 has no per-video model fee because the weights are open. To make the price concrete, use compute pricing: HuggingFace Spaces lists CPU Basic as free, Nvidia T4 small at $0.40/hour, Nvidia L4 at $0.80/hour and Nvidia A10G small at $1.00/hour.
| Path / hardware | Official price | What you get | Source |
|---|---|---|---|
| Register | Free account | Download/access model repositories | Official source ↗ |
| Model weights | $0 model fee | Wan-AI/Wan2.2-TI2V-5B and Wan2.2-I2V-A14B | HuggingFace model card |
| HF Spaces CPU Basic | FREE | 2 vCPU / 16 GB CPU Space | Official source ↗ |
| HF Spaces Nvidia T4 small | $0.40/hour | 16 GB VRAM GPU Space | Official source ↗ |
| HF Spaces Nvidia L4 | $0.80/hour | 24 GB VRAM GPU Space | Official source ↗ |
| HF Spaces Nvidia A10G small | $1.00/hour | 24 GB VRAM GPU Space | Official source ↗ |
05How to prompt
Example: A mountain cabin at blue hour, warm window light flickering, snow drifting slowly across the foreground, gentle dolly-in, quiet cinematic realismOpen Wan model card
06vs the competition
Short version: Wan 2.2 is a strong open local alternative with multiple model sizes. Hailuo 3 / H3 is stronger when REF2VA and first/last-frame control are your main use case.
Read full comparison →07FAQ
Is Wan 2.2 open source?
Yes — Wan 2.2 weights are public on HuggingFace under the Wan-AI organization.
Can Wan 2.2 run locally?
Yes — it runs in ComfyUI and related tooling. VRAM depends on which size you run and how you quantize it; the model cards have current figures.
How much does Wan 2.2 cost?
The open weights are free to download and run yourself. The cost is GPU hardware, cloud GPU rental, or any third-party hosted provider you choose.
Where do I register for Wan 2.2?
For open-weight downloads, create a HuggingFace account at https://huggingface.co/join and use the Wan-AI model pages. Hosted Wan portals may have separate registration.