Model download guide-Cloud Application Platform(CAP)-阿里云帮助中心

ComfyUI is a powerful open-source image generation tool. Its core strength is its support for a wide variety of models. These models determine what ComfyUI can generate and how it generates that content. This guide describes several methods for downloading ComfyUI models. These methods help you obtain the models you need efficiently and conveniently.

1. Model Square: One-stop download for curated models

To simplify how you obtain models and solve the pain point of slow cross-border downloads, we provide Model Square. Model Square collects common ComfyUI models to help you easily find and download what you need.

Download steps:

Log on to the console: Log on to the console and go to your target image generation project.
Access Model Square: In the console, find and click Model Square.
Browse and select: Model Square displays a list of models. Filter and browse by model type and source to find the best model for your needs.
Use the filter bar to filter for models in a specific category, such as Checkpoints.
One-click download: When you find a model you want to use, click the icon in the upper-right corner to add it to My Models. Alternatively, click the model to view its details, then click the Download Model button on the product page to add it.
Click the image to view model details.
Click Download Model. The Add to My Models dialog box appears. It shows the default Model File Name and download Path. You can modify the Model File Name and download Path as needed. Click Confirm to start the download.
After the download starts, a task list automatically appears to show the download progress.
After the download is complete, find the model in the models directory under File Management. You can then use this model in ComfyUI.

2. File Management: Flexibly upload local or cloud models

If you have model files on your local computer or in Object Storage Service (OSS), you can use the File Management feature to upload and integrate them into ComfyUI.

2.1 Upload a model from OSS

Log on to the console: Log on to the console and go to your target image generation project.
Go to File Management: On the project page, click the File Management tab.
Go to the models directory: Go to the subdirectory where you want to upload the model, such as the loras directory, and click Import from OSS.
Select an OSS Bucket and the model file to upload. Click the Import button. The system pulls the model file from OSS and imports it to the selected destination path.

2.2 Upload a model from your local computer

You can upload model files directly from your local computer using the File Management feature.

Log on to the console: Log on to the console and go to your target image generation project.
Go to File Management: On the project page, click the File Management tab.
Select local upload: Go to the subdirectory where you want to upload the model, such as the loras directory, and click Local Upload.
Select the model file: The system opens a file selector. Browse your local file system to find and select the ComfyUI model file that you want to upload.
Confirm the upload: Click the Upload button to upload the model file to your ComfyUI environment.

3. Use the Hugging Face and ModelScope command line tools

If you are familiar with the command line or need the latest models, the most efficient way to download them is using the command line tools for Hugging Face Hub and ModelScope. Both platforms are leading open-source AI model communities.

3.1 Download Hugging Face models using `huggingface_cli`

Hugging Face Hub has tens of thousands of pre-trained models. The official command-line interface (CLI), huggingface_cli, lets you easily download these models.

Prerequisites:

The huggingface_hub Python library must be installed in your environment. If it is not, go to Project Development > Instance List > Log on to Instance and install it by running the following command.

pip install huggingface_hub

Download command:

Use the following command to download a specific model. You can find the full ID for each model on the Hugging Face Hub website.

For more information about downloading models with huggingface-cli, see the Hugging Face documentation.

huggingface-cli download <model_id> --local-dir <local_directory>

<model_id>: The full identifier of the model that you want to download from Hugging Face Hub. For example, the identifier for https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/ae.safetensors is black-forest-labs/FLUX.1-dev ae.safetensors.
--local-dir <local_directory>: The save path on your local file system where the model will be downloaded. Ensure that the path exists and that you have write permissions for it.

Example:

https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/ae.safetensors

Download the ae.safetensors model and save it to the ./models/vae directory:

huggingface-cli download black-forest-labs/FLUX.1-dev ae.safetensors --local-dir ~/comfyui/models/vae/

3.2 Download models using ModelScope

ModelScope is an open-source AI model community led by Alibaba DAMO Academy. It provides a rich collection of Chinese and multi-modal models.

Prerequisites:

The modelscope Python library must be installed in your environment. If it is not, go to Project Development > Instance List > Log on to Instance and install it by running the following command.

pip install modelscope