PAI ArtLab ComfyUI

Use third-party models and custom nodes

1. Ensure that you are using ComfyUI (Exclusive Edition). ComfyUI (Shared Edition) does not support third-party models or custom nodes.

2. Go to the OSS storage directory that is mounted to the workflow. In the upper-right corner, click the icon and click the link to the storage bucket.

   The OSS path has the following directory structure.

   It contains the custom_nodes/, input/, models/, output/, temp/, and unet/ subdirectories.

   • custom_nodes: Stores node files.

     • Example: To install a third-party node, download its library package and upload the folder to the custom_nodes directory. For example, download a folder named ComfyUI_FlashVSR-main and upload it to your custom_nodes directory. On a GitHub repository page, you can get the package by clicking Code > Download ZIP. Decompress the downloaded ZIP file to find the node folder.

   • models: Stores model files.

     • Example 1: Models that a node library depends on

       • Node libraries often depend on specific models. To ensure the nodes run correctly, download and upload the required models. Upload the models to the corresponding directories within the models directory: the FlashVSR directory and the vae directory. All the preceding model files should be uploaded to models/FlashVSR and models/vae.

         ComfyUI/models/FlashVSR
         ├── LQ_proj_in.ckpt                                          # v1.1 or v1.0
         ├── TDecoder.ckpt
         ├── diffusion_pytorch_model_streaming_dmd.safetensors         # v1.1 or v1.0
         ├── posi_prompt.pth
         ComfyUI/models/vae
         ├── Wan2.1_VAE.pth                                           # or safetensors
         ├── lightvae2_1.pth                                          # 32.2M or taev2_1.pth, lighttaev2_1.pth
         ├── Wan2.1_VAE_upscale2x_imageonly_real_v1_diff.safetensors   # rename from diffusion_p

     • Example 2: A workflow reports a missing model error

       • A missing model error usually occurs because the platform cannot connect to the public internet by default. To resolve this, add the required model to the models directory. For example, if depth_anything_v2_vits_fp16.safetensors is missing, add it to the models/depthanything/ directory in OSS. The error message specifies the node that requires the model, such as DownloadAndLoadDepthAnythingV2Model, and the full file path.

       • If the required path does not exist in OSS, create it.

3. Load new models or nodes

   After you upload a model to the mounted storage space, click PaiCustom > Load New Model. If the model is not found, click Restart Process. After the process restarts, refresh the browser page.

   After you upload a node file, click Restart Process. After the process restarts, refresh the browser page.

   If the new model or node is still not found, in the left-side navigation pane, click Developer Center. Find the corresponding ComfyUI service, Stop the service, and then Start it again.

4. Upload model or node files You can upload files to OSS by using the console. For large files, see How do I upload large objects to OSS?.

Export a workflow

Switch UI versions

• Switch from the old UI to the new UI:

  Open the Settings dialog box. In the left pane, select Comfy. In the Region settings area on the right, select New from the drop-down list.

• Switch from the new UI to the old UI:

  Open the Settings dialog box and go to the Comfy section. In the Region settings area, select Old from the drop-down list.

Verify ComfyUI API integration

For more information, see PAI ArtLab Developer Center usage guide.