Documentacao do Wonderful Launchergetting-started
System Requirements
Hardware and software requirements for running ComfyUI.
Hardware Requirements
GPU (Graphics Card)
A dedicated GPU is strongly recommended. ComfyUI can run on CPU, but generation will be very slow.
| GPU Brand | Support Level | Notes |
|---|---|---|
| NVIDIA | Full support | Best experience. RTX 3060+ recommended. CUDA 12.1+ required |
| AMD | Partial | Windows: DirectML. Linux: ROCm. Some nodes may not work |
| Intel Arc | Experimental | XPU support available but not fully stable. See GPU Compatibility |
| Apple Silicon | Good | M1/M2/M3/M4 via MPS backend. macOS only |
| CPU only | Supported | Very slow. Use --cpu flag. Only for testing |
VRAM (GPU Memory)
VRAM determines what models you can run and at what resolution:
| VRAM | What You Can Run |
|---|---|
| 4 GB | SD 1.5 at 512x512 with --lowvram |
| 6 GB | SD 1.5, some SDXL workflows |
| 8 GB | SDXL, Flux (quantized), most image workflows |
| 12 GB+ | Flux (full), video models (short clips) |
| 16 GB+ | HunyuanVideo, Wan 2.1, longer video generation |
| 24 GB+ | Full-resolution video, large batch sizes |
RAM (System Memory)
| Amount | Suitability |
|---|---|
| 8 GB | Minimum. May struggle with large models |
| 16 GB | Good for most workflows |
| 32 GB+ | Recommended for video generation and complex workflows |
Storage
| Space | What It Covers |
|---|---|
| 15 GB | ComfyUI app + Python environment |
| 2-10 GB per model | Checkpoint files (SD 1.5 ~2 GB, SDXL ~7 GB, Flux ~12 GB) |
| 50 GB+ | Recommended total for app + a few models |
Use an SSD for best performance — model loading is significantly faster compared to HDD.
For detailed GPU compatibility across all brands and generations, see GPU Compatibility.
Software Requirements
For Desktop App (Recommended)
- Windows 10 or 11
- NVIDIA GPU with up-to-date drivers
- Stable internet connection (first launch downloads ~15 GB)
- No other software needed — Python is bundled
For Portable Package
- Windows 10 or 11
- 7-Zip to extract the archive (Windows built-in zip may fail)
- NVIDIA GPU drivers (or use CPU mode)
For Manual Install
- Windows 10 or 11
- Python 3.12 or 3.13 (3.13 recommended)
- Git
- NVIDIA CUDA Toolkit (matching your PyTorch version)
- pip (included with Python)
Check Your GPU
Not sure what GPU you have? Open Task Manager → Performance tab → GPU. It shows your GPU model and VRAM.
Or run this in Command Prompt:
nvidia-smi
This shows your NVIDIA driver version and GPU memory. If the command is not found, you may need to install NVIDIA drivers first.
Next Steps
Ready to install? Pick your method:
- Desktop App — recommended for beginners
- Portable Package — no install needed
- Manual Install — full control