How to Download Models from Civitai and Load Them into ComfyUI

A step-by-step guide to installing and using models in ComfyUI for enhanced AI image generation.In this article, i will cover two different ways (from Civitai and from huggingface). These two ways are similar but have some differences.

Introduction

Civitai is the leading platform for sharing AI models . This tutorial will walk you through finding, downloading, and properly loading models from Civitai into ComfyUI, ensuring compatibility and optimal performance.

Prerequisites

Before starting, ensure you have:

• A working installation of ComfyUI (follow the ComfyUI Installation Guide if not already set up)
• Stable internet connection (models can range from 1GB to 10GB+)
• Sufficient storage space (recommend 20GB+ free space for model storage)
• Basic understanding of ComfyUI's filesystem

Before Starting

In this part , i will mainly cover the lora model like z-image-turbo. If you want to download other models like lora, embedding, the method is pretty similar.

Part 1: Downloading Models from Civitai

from Civitai: Access Civitai and Search for Models

1. Open your web browser and navigate to Civitai.com
2. Use the search bar at the top to find models by name (e.g., "Realistic Vision", "DreamShaper") or category.
3. Filter results using the left sidebar:
   • Model Type: Select the model type (Checkpoint, LoRA, ControlNet, Embedding, etc.)
   • Base Model: Choose compatibility (e.g., SD 1.5, SDXL 1.0,they are not compatible even just a few charactors apart. )
   • Read the Description for use cases, strengths, and recommended prompts
   • Review Sample Images to verify the model’s output style
   • Note the File Size to ensure you have enough storage

from Hugging Face: Access Hugging Face and Search for Models

1. Open your web browser and navigate to Hugging Face
2. Use the search bar at the top to find models by name (e.g., "z-image-turbo_fp8") or category.
3. go the files and versions section and choose a model file to download.

• some introduction:

1. you can download vae test_encoders models from huggingface. But not download diffusion models (which is put in the transformers folder) ,because they are not quantized(with big file size and not compatible in comfyui app).
2. download quantifized model provided by community. Find model tree sectioin on the right half of the page. Choose quantified model link, and you will find there are models. Most of them are gguf version.

Part 2: Loading Models into ComfyUI

Step 1: Locate ComfyUI's Model Directories

First, find your ComfyUI installation folder. The default model directories are:

:::tip Organization Tip Create subfolders within each directory (e.g., lora/character/, controlnet/pose/) to keep your models organized as your collection grows. still can be read by the computer correctly. :::

Step 2: Install the Downloaded Model

1. Locate the downloaded model file (usually in your browser’s "Downloads" folder)
2. Copy or move the file to the corresponding ComfyUI model directory (e.g., LoRA file → models/lora/)
3. Do not rename the file unless you understand the implications (some models rely on specific naming conventions)
4. For compressed files (.zip, .rar):
   • Extract the archive first
   • Ensure the extracted file (not the archive itself) is placed in the model directory

Step 3: Restart ComfyUI

1. Close ComfyUI completely
2. Reopen ComfyUI to allow it to detect the new model
3. Wait for ComfyUI to load—model scanning happens automatically on startup

Part 4: Troubleshooting Common Issues

Issue 1: Model Not Appearing in ComfyUI

• Verify the model is in the correct directory (e.g., LoRA in models/lora/, not models/checkpoints/)
• Ensure the file extension is supported (.safetensors)
• Restart ComfyUI again (sometimes models take two restarts to appear)
• Check that the file wasn’t corrupted during download (re-download if necessary)

Issue 2: Model Causes ComfyUI to Crash

• The model may be incompatible with your ComfyUI version (update ComfyUI)
• Insufficient VRAM: Use smaller models (e.g., pruned versions) or enable model offloading in ComfyUI settings
• Corrupted file: Re-download the model from Civitai
• Base model mismatch (e.g., using an SDXL model with an SD 1.5 workflow)

Issue 3: Poor Quality Output

• Use the recommended prompts from the Civitai model page
• Adjust model strength (for LoRAs/ControlNets)
• Ensure you’re using the correct VAE (some models require specific VAEs for optimal results)
• Check for model version compatibility (e.g., SDXL 0.9 models vs. SDXL 1.0)

Issue 4: Slow Generation Speed

• Enable model offloading in ComfyUI settings (Settings → Performance → Enable Model Offloading)
• Use smaller resolution or batch sizes
• Consider using model quantization (e.g., 4bit/8bit) for large models
• Upgrade your GPU VRAM (long-term solution for handling large models)