Run ComfyUI on a rented GPU. Choose a HearmemanAI template, select your models and deploy.
Choose a template
- Wan for Wan video workflows.
- Qwen / Krea / Z Image for image workflows.
- MiniMax H3 for H3 video and audio workflows.
- LTX 2.5 / 2.3 for LTX video workflows.
Open a template and click Deploy on Runpod. Check the selected template before continuing.
Pick a GPU
Match the GPU's VRAM to your workflow, resolution and model precision. Check the template's requirements and rental price before deploying.
Enable your models before deployment
- Click Edit Template before deploying.
- Expand the environment variables section.
- Set the model families you need from
falsetotrue. MiniMax H3 and LTX 2.3 are already enabled. Leave unused families off. - Save the template settings and deploy. First boot downloads the models you enabled, so give the downloads time to finish.
- Open ComfyUI on port
8188. JupyterLab is available on8888when you need access to files.
Download only what you need
Enabled models download directly to the pod. More models mean longer startup times and more disk space.
Open the workflow
Look in ComfyUI's workflows menu for the workflows bundled with your template. To import a separate file, follow Load a workflow JSON.
For a CivitAI LoRA or checkpoint that is not covered by a model flag, use the template's supported downloader. See Download CivitAI models at boot.
Finish your session
Save your outputs and stop the pod when you finish. Storage charges may continue separately.
If the workspace does not open, use Troubleshoot a Runpod workspace.