ComfyUI starter

The information provided here includes things I found helpful for getting started.

This starter is intended as a supplement to the existing ComfyUI README.md, which you can find here: ComfyUI GitHub

It assumes that you have already successfully installed ComfyUI.

Key Terms

• Checkpoint (Base Model): The main AI model containing learned weights (like SD1.5). Start your workflow here.
• LoRA (Low-Rank Adaptation): Small files adjusting the style or subject of your base model. Requires a checkpoint.
• VAE (Variational Autoencoder): Handles image encoding/decoding; affects image quality and colors.
• Workflow: A visual graph of nodes that instruct ComfyUI on generating images.

Setting Up Models

• Checkpoints: Place .safetensors or .ckpt files in ComfyUI/models/checkpoints.
• LoRA: Place LoRA files (.safetensors) in ComfyUI/models/loras.
• VAE: Optional. Place VAE files (.safetensors) in ComfyUI/models/vae.

Using ComfyUI Manager

ComfyUI Manager simplifies managing extensions and nodes.

• To install manually:

  git clone https://github.com/ltdrdata/ComfyUI-Manager custom_nodes/comfyui-manager

• Restart ComfyUI, and use the built-in Manager interface to browse and install nodes.

Building a Workflow

Basic text-to-image workflow:

1. Load Checkpoint Node → select your base model.
2. CLIP Text Encode Node → input your prompt.
3. KSampler Node → generates the latent image.
4. VAE Decode Node → converts latents into visible images.
5. Save Image Node → saves the generated images.

Using LoRAs:

• Add a Load LoRA Node after your checkpoint.
• Set LoRA strength to control its influence.

Finding & Using Workflows

• Workflows can be saved/loaded as .json files or embedded in image metadata.
• Drag workflow images into ComfyUI to instantly load setups.
• Official workflows and examples: ComfyUI GitHub

Desktop Version

• Easy setup with built-in dependencies, automatic updates, and a new intuitive UI.
• Includes ComfyUI Manager by default.
• Ideal for beginners: Download here

happy generating!