Running DeepSeek Janus Pro 1B on Windows with ComfyUI: Step-by-Step Guide

Anas Mohammad -

Running DeepSeek Janus Pro 1B on Windows using ComfyUI is an exciting venture into the world of multimodal AI models. This guide provides a comprehensive, step-by-step approach to setting up and utilizing this powerful model on your Windows machine.

We cover everything from prerequisites to installation, configuration, and practical usage examples, ensuring you have a thorough understanding of the entire process.

Why DeepSeek Janus Pro 1B?

Key Benefits:

  • 🎨 Text-to-Image Generation: Transform prompts into stunning visuals
  • 🖥️ Local Installation: No cloud costs or internet required
  • 🚀 GPU Acceleration: NVIDIA CUDA support for fast results
  • 🔓 Open Source Model: Fully customizable for developers

System Requirements

Component Minimum Spec Recommended Spec
OS Windows 10 64-bit Windows 11 64-bit
CPU Intel i5 8th Gen Intel i7/Ryzen 7
RAM 8GB 16GB+
GPU Integrated NVIDIA RTX 3060 (8GB VRAM)
Storage 10GB SSD 20GB NVMe SSD

Step-by-Step Installation Guide

1. Install Core Dependencies

Python 3.10 Installation:

  1. Download from python.org
  2. Check "Add Python to PATH" during installation

Verify in Command Prompt:

python --version

CUDA Setup for NVIDIA Users:

  1. Check compatible CUDA version for your GPU
  2. Download from NVIDIA CUDA Toolkit
  3. Install with default settings

2. Set Up ComfyUI Environment

# Clone official repository
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

# Create virtual environment
python -m venv venv
venv\Scripts\activate

# Install core requirements
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

3. Install Janus Pro Extension

cd custom_nodes
git clone https://github.com/CY-CHENYUE/ComfyUI-Janus-Pro.git
pip install -r ComfyUI-Janus-Pro/requirements.txt

4. Download Model Files

  1. Visit Hugging Face Model Hub
  2. Download:
    • config.json
    • pytorch_model.bin
    • tokenizer.json

Create folder structure:

ComfyUI
└── models
    └── Janus-Pro
        └── Janus-Pro-1B
            ├── config.json
            ├── pytorch_model.bin
            └── tokenizer.json

Launching ComfyUI: First-Time Setup

cd ComfyUI
python app.py --listen

Access the UI at http://localhost:8188 in your browser.

Generating Your First Image

  1. In ComfyUI, right-click workspace → Add Node
  2. Select Janus Pro Loader from the menu
  3. Add Text Prompt node and connect to loader
  4. Configure parameters:
    • Prompt: "A cyberpunk city at sunset"
    • Negative Prompt: "blurry, low quality"
    • Dimensions: 1024x768
  5. Click Queue Prompt to generate

Advanced Configuration Tips

Optimizing Performance

Scenario Solution
Low VRAM GPUs Use --lowvram launch flag
CPU-only Mode Set "device": "cpu" in config
Batch Processing Adjust num_images parameter

Custom Workflow Example

{
  "prompt": {
    "text": "A steampunk airship in the clouds",
    "negative_prompt": "modern elements",
    "steps": 50,
    "cfg_scale": 7.5,
    "seed": 42
  },
  "output": {
    "resolution": [1152, 896],
    "format": "PNG",
    "quality": 95
  }
}

Troubleshooting Common Issues

Problem: CUDA Out of Memory Error
Fix: Reduce batch size/image dimensions

Problem: Missing Model Files
Fix: Verify folder structure matches:

ComfyUI/models/Janus-Pro/Janus-Pro-1B/

Problem: Slow Generation Speed
Fix: Enable xFormers in launch command:

python app.py --xformers

Pro Tips for Better Results

  1. Use descriptive prompts: "4k detailed portrait" vs "a person"
  2. Experiment with negative prompts to remove unwanted elements
  3. Adjust CFG scale (7-12 works best for most scenarios)
  4. Use seed values for reproducible results

Beyond Basics: What's Next?

  1. Animation Workflows: Combine with AnimateDiff
  2. Upscaling: Integrate ESRGAN models
  3. ControlNet: Add pose/edge detection
  4. Custom Training: Fine-tune with your dataset

Conclusion

Setting up DeepSeek Janus Pro 1B on Windows using ComfyUI opens up numerous possibilities for exploring AI-generated content. Whether you're a developer looking to integrate AI into applications or a researcher exploring new frontiers in machine learning, this setup provides a robust foundation for your projects.