ComfyUI lesson 1: Install ComfyUI cleanly and make your first image
Pick a portable install, verify the GPU, place one checkpoint correctly, and run a known-good graph.
Listen to this article
On-device voiceUses the voice built into your browser; no article text leaves this page.
Audio narration is not supported by this browser.
ComfyUI becomes much easier once the installation is boring. This lesson gets one known checkpoint, one known workflow, and one successful output working before custom nodes or giant model packs enter the picture.
Pick an install you can repair
On Windows with an NVIDIA GPU, the official portable build is the least surprising starting point. Extract it into a short path such as C:\AI\ComfyUI; do not run it from inside the archive or a cloud-synced folder. On Linux, use a virtual environment and install the PyTorch build matching the CUDA runtime supported by your driver. Apple Silicon uses the normal Python install with MPS, but many CUDA custom nodes will not apply.
Before adding anything, launch ComfyUI and read the first 30 lines of the terminal. You want to see the detected device and VRAM, not a silent CPU fallback. Keep that terminal open: the browser shows the graph, while the terminal usually explains the real error.
Know the four folders
| Asset | Folder |
|---|---|
| Full checkpoint | models/checkpoints |
| LoRA | models/loras |
| ControlNet | models/controlnet |
| VAE | models/vae |
Download Proteus V0.4 from its linked model source and keep the gallery filename proteusV0.4.safetensors. A workflow references filenames, not your intention; renaming a weight is enough to make an otherwise correct graph fail.
Your first controlled run
Import the lesson JSON, refresh models, select the checkpoint if needed, and queue the graph without changing anything. The graph is deliberately ordinary: checkpoint loader → two text encoders → empty latent → sampler → VAE decode → save.
If it succeeds, change only the prompt. Then change only the seed. Save the output and the workflow together. You now have a baseline that can distinguish an installation problem from a creative experiment.
Do not install a node manager pack, 40 extensions, and six checkpoints yet. First prove that core nodes, one model, the GPU, and the output directory agree.
Run the lesson, do not just read it
- Open the tested workflow in Comfy Studio
- Download the exact API-format JSON
- CLI:
app comfy download proteus-v04-text-to-image
The gallery records model sources, minimum VRAM, custom-node requirements, and verification state beside the graph. That is the portable unit for this series.
Series progress: lesson 1 of 10.