ComfyUI looks for checkpoint files in /workspace instead of /runpod-volume

[plush willow]

I had a comfyUI on-demande GPU pod, and now need to switch a serverless pod. After setting up the endpoint, I can run some requests, but I see my comfyUI workflow says there are missing checkpoints and LoRa. My serverless workers are correctly connected to my 100Go volume. So it seems the path is actually different in both instances.

How can I either:

• move the files from /workspace/comfyUi/checkpoints to /runpod-volume/comfyUI/checkpoints ?
  or
• have comfyUI load the files from a different directory?

thanks ❤️

[lone bone]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[plush willow]

For context i've been following this guide here on the comfyUI worker image README:

But when triggered with the APi teh workslow says there is only the base flux model

[plush willow]

I downloaded the comfyui image on github, added the downloads to the Dockerfile.. I can see the downloads going through all well in the logs. But then at the very end when the server is testing, it fails again with this:

So even tho im following the customisation steps for the comfyui image, its not saving in the right place 🤷 ?

[loud venture]

network storage on serverless is always mount at same path

[plush willow]

Thanks for your answer ! The issue is that if you ls inside the on-demande GPU pod's /workspace you see the ComfyUI folder, which is not the correct path to access through serverless.

If you look in the first reply to this thread (cf screenshot): https://www.answeroverflow.com/m/1249796140283727872

It seem you are correct in saying its always the same mount path, but why wouldnt my image find the models even after they are uploaded?

Should I change the relative path to download the customs model in my Dockerfile?

[loud venture]

are you sure you need relative paths comfy-cli is smart and should place files in correct folder when downloading from civaticai

though in this docker file you embed models into docker image

so no point to use network storage

[plush willow]

Yes i tried both methods but I get the same error everytime, only the flux model available 😭

[plush willow]

let me try the dockerfile without relative path

They still need to know if a donwload goes into the checkpoint or lora folder tho

Or you advise to remove the --relative-path flag completely ?

trying now with explicit --dest like:

# Download the main checkpoint model
RUN comfy model download \
    --url "https://civitai.com/api/download/models/1719716?type=Model&format=SafeTensor&size=full&fp=fp16" \
    --dest /runpod-volume/ComfyUI/models/checkpoints \
    --filename "animicsPonyXL_v50.safetensors"

[plush willow]

I think the custom image worked and its just failing the tests because the tests are expecting the base flux model.. Is it ok to just delete all the /tests workflows?

[loud venture]

you using github integration?