**In this blog i will show you how to deploy Wan2.1 model in your own machine easily even if you are noob developer. I will explain everything step-by-step with simple language. Let’s start!**
*Table of content:*
– What is Wan2.1 model
– Why use Wan2.1 (5 reasons)
– Prerequisites needed (hardware/software)
– How to install Wan2.1 (step-by-step)
– How to run text-to-video (with examples)
– How to use image-to-video feature
– How to fix common errors (OOM, dependency issues, etc)
– Tips for better results
– Conclusion & what’s next
—
### **What is Wan2.1 model**
Wan2.1 is **free open-source AI model** made for video generation. Think of it like ChatGPT but for videos! You give text or image, it makes video. Here key features:
– **Text-to-Video**: Type “dog dancing in rain”, get video.
– **Image-to-Video**: Upload cat photo → video of cat jumping.
– **Supports 480P/720P**: Good quality for social media.
– **Works on normal GPUs**: No need expensive servers. RTX 3060/4090 enough.
– **Adds text in videos**: Like subtitles or memes (supports English/Chinese).
– **Free forever**: Unlike Midjourney or Sora, no paid plans.
*Fun fact*: Wan2.1 can make 5-second video in 4 mins on RTX 4090. Slower than TikTok filters but way more powerful!
—
### **Why use Wan2.1? (5 reasons)**
1. **No coding skills needed**: Just copy-paste commands.
2. **Good for small projects**: Make YouTube shorts, ads, memes.
3. **Learn AI cheaply**: Great for students or hobbyists.
4. **Privacy**: Run locally, no data sent to cloud.
5. **Customizable**: Want cat with blue hat? Just type it!
*Example use cases*:
– YouTubers: Create intro/outro videos.
– Teachers: Make educational animations.
– Meme creators: Turn jokes into videos.
—
### **Prerequisites needed**
Before installing, check if your PC meets these:
#### *Hardware*:
– **GPU**: **NVIDIA with 8GB+ VRAM** (RTX 3060/Ti, 4090 etc).
– AMD GPUs not recommended (issues with PyTorch).
– **RAM**: 16GB+ (32GB better for 14B model).
– **Disk space**: 40GB+ (models are BIG).
#### *Software*:
1. **Python 3.10+**: Download from [python.org](https://python.org).
2. **Git**: For cloning repo. Install from [git-scm.com](https://git-scm.com).
3. **Pip**: Usually comes with Python.
*Note for Windows users*:
– Enable “Developer Mode” for Linux-like commands.
– Use **CMD/PowerShell**, not Notepad!
—
### **How to install Wan2.1 (step-by-step)**
Follow these steps carefully. If stuck, comment below!
#### Step 1: Clone GitHub repo
Open terminal (Linux/Mac) or CMD (Windows) and type:
“`
git clone https://github.com/Wan-Video/Wan2.1.git
cd Wan2.1
“`
*This downloads code to your PC*.
#### Step 2: Install dependencies
Type:
“`
pip install -r requirements.txt
“`
Wait 5-10 mins. If you see red errors:
– **Fix missing packages**: Run `pip install [package-name]`.
– **Update pip**: `pip install –upgrade pip`.
#### Step 3: Download model
Choose model based on your GPU:
| Model | VRAM Needed | Quality | Disk Space |
|——-|————-|———|————|
| 1.3B | 8GB | Good | 15GB |
| 14B | 24GB | Best | 30GB |
For 1.3B (beginners):
“`
pip install huggingface_hub
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B –local-dir ./Wan2.1-T2V-1.3B
“`
For 14B (high-end GPUs):
“`
huggingface-cli download Wan-AI/Wan2.1-T2V-14B –local-dir ./Wan2.1-T2V-14B
“`
*Tips*:
– If download slow, use **Modelscope** (link in GitHub).
– Use `–resume-download` if internet breaks.
—
### **How to run text-to-video (with examples)**
Now the fun part! Let’s make your first video.
#### Basic command (1.3B model):
“`
python generate.py –task t2v-1.3B –size 832×480 –ckpt_dir ./Wan2.1-T2V-1.3B –prompt “cute robot dancing in disco lights”
“`
– `–task t2v-1.3B`: Use 1.3B model for text-to-video.
– `–size 832×480`: Video resolution (480P).
– `–ckpt_dir`: Path to downloaded model.
– `–prompt`: Your video idea (English or Chinese).
Wait 5-10 mins. Video saves in `outputs` folder.
#### Advanced example (14B model):
“`
python generate.py –task t2v-14B –size 1280×720 –ckpt_dir ./Wan2.1-T2V-14B –prompt “a spaceship flying over Mars, 4K cinematic, dramatic music” –seed 42
“`
– `–seed 42`: Fix random number for same output every time.
– `–size 1280×720`: 720P HD video.
#### If you get CUDA memory error:
Add `–offload_model True –t5_cpu` to free GPU memory:
“`
python generate.py –task t2v-1.3B –size 832×480 –ckpt_dir ./Wan2.1-T2V-1.3B –offload_model True –t5_cpu –prompt “panda eating noodles”
“`
—
### **How to use image-to-video feature**
Wan2.1 can turn images into videos. Example: Make a photo of flower “bloom” in video.
#### Step 1: Prepare image
– Save image as `input.jpg` in `Wan2.1/examples/` folder.
#### Step 2: Run command
“`
python generate.py –task i2v-14B –size 1280×720 –ckpt_dir ./Wan2.1-I2V-14B-720P –image examples/input.jpg –prompt “red rose blooming in sunlight, slow motion”
“`
– `–task i2v-14B`: Image-to-video task.
– `–image`: Path to your image.
*Note*: For 480P video, use `Wan2.1-I2V-14B-480P` model.
—
### **How to fix common errors**
#### 1. **CUDA out of memory**:
– **Fix 1**: Reduce video size (use 480P instead of 720P).
– **Fix 2**: Add `–offload_model True –t5_cpu`.
– **Fix 3**: Close Chrome/other apps using GPU.
#### 2. **Dependency errors**:
“`
pip install –upgrade torch numpy transformers
“`
#### 3. **Model not found**:
Check `–ckpt_dir` path. Folder names must match:
– ✅ Correct: `./Wan2.1-T2V-1.3B`
– ❌ Wrong: `./wan21-1.3b`
#### 4. **Black screen video**:
– Change seed value: `–seed 123`
– Use simpler prompts (e.g., “cat sitting” vs “cat flying in space with laser eyes”).
—
### **Tips for better results**
1. **Use detailed prompts**:
– Bad: “car racing”.
– Good: “4K ultra-realistic Ferrari racing on wet road, dramatic camera angles, rain splashing, cinematic lighting”.
2. **Adjust `–sample_guide_scale`**:
– Higher values (6-10) = more creative.
– Lower values (3-5) = follow prompt strictly.
3. **Use **prompt extension**:
Add `–use_prompt_extend` to auto-improve your prompt:
“`
python generate.py … –use_prompt_extend
“`
4. **Start small**: Test 2-second videos before making 10-second ones.
—
### **Conclusion & what’s next**
You’re now ready to make AI videos! Key takeaways:
– Use **1.3B model** if GPU is weak.
– **Image-to-video** works best with 720P model.
– Join **Discord/WeChat** groups (links in GitHub) for help.
*What’s coming next*:
– **Wan2.1 for ComfyUI** (drag-drop interface).
– **Mobile app** (run on Android via Termux).
Try making a video and share results in comments! 😊
—
**FAQs**
Q: Can I use CPU instead of GPU?
A: No, too slow. Even 10 sec video will take 1 day.
Q: How to make longer videos?
A: Not supported yet. Max 5-10 sec for now.
Q: My video has weird colors?
A: Update PyTorch and CUDA drivers.
Q: Can I train my own model?
A: Yes, but need 4x A100 GPUs (not for beginners).
—
*Need more guides? Check my blogs on [Host Python apps free](…) and [Run code in mobile browser](…).*
