ComfyUI is the most powerful node-based interface for running local AI image and video generation. This guide gets you from zero to a working workflow on Windows with an NVIDIA GPU.
Hardware Requirements
| GPU | VRAM | Recommended Use |
|---|---|---|
| RTX 5080 | 16GB | Full quality, all models |
| RTX 4090 | 24GB | Full quality, batch generation |
| RTX 3080 16GB | 16GB | Full quality, slower |
| RTX 3080 10GB | 10GB | Reduced resolution |
| RTX 3060 | 12GB | Standard models only |
Step 1 — Install Prerequisites
Python 3.10.x (required)
Download from python.org — use exactly 3.10.x, not 3.11 or 3.12.bash
python --version
Git
bash
git --version
CUDA Toolkit
Download CUDA 12.1 from nvidia.com/cuda-downloads. Match your driver version.bash
nvcc --version
Step 2 — Clone ComfyUI
bash
cd C:\ git clone https://github.com/comfyanonymous/ComfyUI cd ComfyUI
Step 3 — Install Dependencies
bash
python -m venv venv.\venv\Scripts\activate
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt
Step 4 — Install ComfyUI Manager
ComfyUI Manager lets you install custom nodes with one click.
bash
cd custom_nodes git clone https://github.com/ltdrdata/ComfyUI-Manager cd ..
Step 5 — Download Your First Model
Place models in ComfyUI/models/checkpoints/
bash
Step 6 — Launch ComfyUI
bash
python main.pypython main.py --gpu-only --highvram
python main.py --lowvram
Open your browser to: http://127.0.0.1:8188
RTX 5080 Optimal Settings
These settings maximize quality and speed on 16GB VRAM:
Batch size: 2–4 images Resolution: 1024x1024 (SDXL) / 768x512 (video) Steps: 20–30 CFG Scale: 7.0 (images) / 3.0 (video) Sampler: dpmpp_2m Scheduler: karras VAE: Built-in (bake into checkpoint) Precision: fp16
Installing Essential Custom Nodes
In ComfyUI Manager, install these nodes first:
- 1ComfyUI-Impact-Pack — face detailing, segmentation
- 2ComfyUI_IPAdapter_plus — image prompt adapter
- 3ComfyUI-AnimateDiff-Evolved — video animation
- 4ComfyUI-VideoHelperSuite — video input/output
- 5rgthree-comfy — better node organization
Your First Workflow
Once launched, load the default workflow:
- 1Click Load in the top menu
- 2Select
default_workflow.json - 3Change the checkpoint to your downloaded model
- 4Click Queue Prompt
Performance Benchmarks
Tested on RTX 5080 16GB:
| Model | Resolution | Steps | Time |
|---|---|---|---|
| SDXL | 1024x1024 | 20 | ~3.2s |
| FLUX Dev FP8 | 1024x1024 | 20 | ~8.1s |
| SD 1.5 | 512x512 | 20 | ~0.8s |
| LTX Video | 768x512 97f | 25 | ~5.4s |
Common Issues
"CUDA out of memory" — Add--lowvram to launch command, or reduce batch size to 1.
Black images — Wrong VAE. Download the correct VAE for your model.
Slow generation — Ensure --gpu-only flag is set and venv is activated.
Nodes missing — Open ComfyUI Manager and click "Install Missing Custom Nodes".