Runpod installation of Wan2.1 image2video
Log into Runpod and under 'Storage' create a network volume, 100GB is a good size
Make sure the datacenter you select has the GPU you want, e.g. Romania (EU-RO-1)
You need a 4090 or 5090 or another NVIDIA card with at least 24GB of VRAM
On your new storage drive, click 'deploy'. It might take a few seconds for the new storage drive to show up in the 'storage' section
On the Deploy page, choose the GPU you want
Choose this Pod template: 'ComfyUI with Manager inst. Permanent Disk torch2.4'
Make sure Jupyter Notebook is enabled and click 'Deploy'
You will be redirected to the GPU Pods list
Expand the pod details (down arrow on the right) and wait for the 'connect' button to light up
On the Connect page, wait for Jupyter to become available, it will take 10-15 mins
If you get bored you can use the Logs button to track progress
Once Jupyter shows 'ready', launch it, and then click on the 'Terminal' button
First we'll download the models to the persistent storage drive. This only needs to be done once, as long as you keep the network volume:
Choose a wan2.1 model, if you want the 480p image2video use this:
And if you want 720p use this: (You can download both if you want to experiment)
Also get the rest of the files:
This part is required for the 5090. The CUDA bit is installed to the OS disk so needs to be done again you create a new pod, but the Comfy update only needs to be done once since Comfy is on the persistent drive.
Make sure it says Build cuda_12.8 then you can start comfyui:
This takes a couple of minutes but when it's done go back to runpod, click connect, and click on HTTP Service once it turns green
ComfyUI will launch
on the main Comfy screen, click the blue 'manager' button and then hit 'update comfyui'
when the update is complete it will restart comfy
Now open the workflow by downloading this JSON file and dragging it into ComfyUI
Upload an image of exactly the right size (e.g. 480x832), set the prompt, adjust the video X and Y and frames, and hit generate!
When you are finished with the pod, go back to Runpod and choose the trashcan (terminate pod)
Your gens will not be deleted and will stay on the persistent storage for next time. The 100GB storage costs about $7 per month. If you're not going to use it for a week or more it's cheaper to just delete the storage pod and start from scratch next time.
Re-using the network volume
Log into Runpod and go to the Storage section again. On your network volume, click 'Deploy'.
Choose the GPU, make sure 'ComfyUI with Manager inst. Permanent Disk torch2.4' is still selected in the Pod Template section, and click 'Deploy'
You will be redirected the the GPU Pods list
Expand the pod details and wait for the 'connect' button to light up
On the Connect page, wait for Jupyter to become available, it will take 1-2 mins
In Jupyter click on Terminal and install CUDA 12.8 again if you're using a 5090:
This only takes 2-3 minutes and then you can start ComfyUI
Now go back to runpod, click connect, and click on HTTP Service once it turns green, and ComfyUI will launch. You do not need to update ComfyUI.
LoRA usage
If you want to add a LoRA from civitai, enable the Load Lora node in the workflow
The LoRA file has to be placed in the ComfyUI/models/loras folder and you can do this in Jupyter after downloading the file from Civit. Navigate to the loras folder and drag the lora file into Jupyter.