Custom checkpoint on Comfy UI Serverless
What is the best way to use a custom checkpoint/loras/custom nodes on a serverless comfyUI?
WDownload it on your machine, build your dockerfile with the file in it
WDon't use network volume unless you want really SLOW checkpoint loading speed
ABut the image will be like very heavy
WYou don't care, as long as you put at least 2 max workers, you will always have one worker ready
WYou don't pay for the size of your docker nor the max workers set. Only the active workers. so don't worry about the size
WI usually have 40GB+ of models in my serverless dockers. But be careful to not hit the 80GB max size
WI find that 45GB-50GB is the limit for the models (if you have ComfyUI and custom nodes installed)
AI would be interested to chat with you about that ! I'm new to this
WI'm not part of the team, just a customer, but surely
WI have some free time now as I'm filling a ticket because I'm heavily confident something is not working on their side
ASo I'm trying to use a custom checkpoint (downloaded on my pc). I've created the image and push it on Docker Hub. I use it in runpod, but when I run the request it tells me that the checkpoint does not exists
AMy Dockerfile simply look like this :
FROM runpod/worker-comfyui:5.4.1-base
COPY ./checkpoints/checkpoint.safetensors /workspace/ComfyUI/models/checkpoints/checkpoint.safetensors
RUN ls -lah /workspace/ComfyUI/models/checkpoints || true
FROM runpod/worker-comfyui:5.4.1-base
COPY ./checkpoints/checkpoint.safetensors /workspace/ComfyUI/models/checkpoints/checkpoint.safetensors
RUN ls -lah /workspace/ComfyUI/models/checkpoints || true
WThe image exists locally ?
WYou've built it ?
AYes I built it and push it to Docker Hub, made it private and added the PTA on Runpod
WCan you do this ?
FROM runpod/worker-comfyui:5.4.1-base
RUN ls -lah /
FROM runpod/worker-comfyui:5.4.1-base
RUN ls -lah /
WI think /comfyui is at the root no ?
Wnot in a /workspace
ALike write it in Dockerfile and try to built it with that ?
WCreate a dockerfile with
Build it locally and run it, and check if a /workspace folder exists
Build it locally and run it, and check if a /workspace folder exists
AHow can I see that after I build it ?
Wdocker build -t test-comfyui .Wdocker run --rm -it test-comfyui /bin/bashWyou should have access to the docker and type anything you want
Wsuch as
ls -lah /Wor you can execute the ls command directly with
docker run --rm test-comfyui ls -lah /Wdoes it work ?
AI'm trying it !
AI'm having this
A
Wsee, you have no /workspace
Wcomfyui is at the root
WWthis should be the command
AOh okay !
AMake sense
Wrebuild your docker and check that you correctly have the file
WWand then you can
lsyou'll see if your checkpoint has been successfully copied
WIf you don't have NVIDIA Gpu, or didn't run your docker via docker compose it would link your NVIDIA GPU to your docker automatically,
W
AI have a NVIDIA GPU I think it does not have any effect !
Wthats' a docker compose example. It integrates the correct environment variables to make your GPU compatible with your docker
you can build it with
run it with
you can build it with
run it with
AI'm building it it takes a moment 
Wyou're building it via docker compose build ?
ALike : docker compose build -t comfy-worker:my-model . ?
AWhat is that exactly ?
Wjust docker compose build
