Hey folks, I’m planning to update the installation files with the latest features. It would be great if you could suggest some of the latest, reliable extensions and models. I was considering including FLUX, but I think it might be too large and costly for most beginners. So, I’m thinking of providing two versions of the installation files: a standard minimal version (16GB RAM, 120GB storage) and a more advanced version (32GB RAM, 200GB storage). Let me know your thoughts!
- Flux shcnell is less large then flux dev no? - Can you keep the simple one and make another version? I like to have simple things to learn, but maybe it is best to upgrade? your call. - I would like to learn how to "control" who gets access. I dont want any person with the ip to be able to enter, but I dont want also to have only one ip to be able. There must be something else that can say if someone can enter or not, I dont knwo yet. - Tell me, in that CLI black window I could theoreticaly create a python script of my choice that can itself control the comfy right? - question what if you want to gather mulriple prompt thourgh the day, and only want the server to run them all at once at the end of the day, would you PAY for the server beign run despite not using the GPU through the day (just gathering prompts)? - sorry for many questions, I am excited about this topic
@@EternalAI-v9b Yup, it is configurable but most of the ways require some extra steps in terminal so I avoid sharing them. There is 1 feasible-looking way i want to check: "API Gateway or Reverse Proxy with Token Authentication: If you're exposing certain services via HTTP/HTTPS (e.g., a web server on the VM), you can front those services with an API gateway or reverse proxy that enforces authentication via dynamically generated tokens or OAuth. Steps: Set up an API gateway (such as Kong, NGINX with JWT, etc.). Use a service like OAuth2 or JWT to generate tokens for trusted users. Configure the reverse proxy to validate these tokens before allowing traffic through." Through this we might create a password and only the ones with password can access.
I would definitely include FLUX dev, I see it being used a lot for photorealism. Also, it would be great to let us know what happens if we update and change the GPU to add some faster ones and increase the memory which would lead to changing the location. For the most part, you can't do much with basic libraries and at some point, you have to agree that you have to spend more if you want to achieve good results.
@ns_the_one Flux has GGUF and many other versions that you can use for a tutorial. something that's more interesting to me (and possibly to a lot of people) is deploying ComfyUi as API endpoint on GCP. Do you know how we can do that, so we run our workflows as API endpoints in the cloud? This way we can create SaaS apps with GCP hosted comfyui backend
Using this video as inspiration, I set up a g2-standard-12 instance with the L4 processor; it's 4x the speed of the T4. Costs more per hour, but still less than $1, and with 24GB of VRAM, it's capable of running the full hunyuan video model, and perhaps even Nvidia Cosmos. Note that you can save a few cents and increase security by not reserving a static IP address. Instead, create a loopback by adding --ssh-flag="-L 8188:localhost:8188" to the "gcloud compute ssh" command, and then just connect to localhost:8188 on your local machine.
Oh my gosh this actually worked. Not gonna lie, I nearly gave up when you got to the Terminal part because I know absolutely nothing about code and have only used Terminal once or twice because I find it so intimidating but I pushed through and now it's up and running on my intel mac! Didn't think it was possible to use ComfyUI, or any stable diffusion on this computer - thank you!
wow very nice... you're the superman bro. Thank you! I read in comments you would be making another video for comfyUI with GCP as API backend... can't wait to see it.
@@ns_the_one You are a legend because I just searched "install comfy ui on your own sever" and was afraid because I didnot finf any good result within the 3 first results, thankfully I found yours. I will watch it now. What I liked the most is the sentece you wrote on your video descripttion about its cost. I really like that. I did not watch the video fully yet, but can we customize the server and add our own pyton scripts etc? I wonder. Will watch and see
@@EternalAI-v9b haha thanks, i just youtubed "install comfy ui on your own sever" couldnt find my video as well. Yes you can customize it. This tutorial, in it's essence, is showing you how to create a virtual machine on cloud -with GPU support- and use it for random stuff.
Hey guys, there was a small error in the code which i thought i fixed it but it wasnt fixed. Well, this time it is fixed for sure! Everything should be working well without any issues by now. Still, please leave a comment if you encounter any problems.
@@entertainmentchannel9632 You mean GPU quota? I can suggest to change the text in the explanation field. Write there some unique text about experimenting with GPUs. Also which country are you accessing the GCP from ?
Thanks for creating this💪. I just made it work but total installation took around 1 hour. And as far as I understand, if you create a new GCP account, storage and CPU cost will consume your free credits and wont cost you extra.
Question about authentificaiton (13:19), I read somewhere that if you create multiple google account and multiple google cloud accounts with the same credit card, gogole mmight terminate the account becuase you are trying to get the free tier in duplicate perhaps. In this case I see you using 2 accounts, did you choose the one where you had a google cloud account, and left the other account where you dont have any DEV account or whatsoever? Curious. Maybe I am wrong
This is amazing, thank you! Do you think that also the rendering of AI Videos could be suffraged onto a cloud server? I would be super curious about a material about it!
if I wanted to install Automatic1111 & Fooocus alongside your setup, anything I should keep in mind as to not break anything and be able to import/use the same models ?
Wow thanks! This is what I was looking for! How can I access the output images folder? and is there a workflow to upload them to a service like Google Drive to download them on my Mac?
Thx for the reminding me. I just updated the readme file accordingly. Here you go : 1. Find out the full path of your ComfyUI output folder. Mine is `/home/ns/ComfyUI/output`. 2. Create a new folder in your local machine and cd into that directory through terminal 3. run this command (change "ns" with your comfyvm username and also change the output directory path accordingly ) ```gcloud compute scp --recurse ns@comfyvm:/home/ns/ComfyUI/output ./``` This will download the images into your local machine.
@@ns_the_one - will it also delete them? Or due you have another terminal command to --recurse, then delete. I mainly want to do large batches and upscales, to be refined in Photoshop, so I see no reason to keep them on GDrive after downloading. BTW: great video and perfect timing for me. Moving to the next stage of AI-gen and a. Mac is the worst for this, but best for everything else; b. I don't want to maintain a 'Doze PC/4090 (or even Linux) just for AI. CloudGPUs are the way to go. EDIT: ChatGPT just gave me this code to delete if anyone is wondering: "gdrive list | grep | awk '{print $1}' | xargs -I {} gdrive delete {}"
@@TheDocPixel The code you show is about gdrive. The code i provided is to download images to your local folder. Once you download images you can simply run gcloud compute ssh comfyvm --command "rm -rf ComfyUI/output/*" to delete the images inside the output folder of comfyui
HI, you told me about these steps: "Set up an API gateway (such as Kong, NGINX with JWT, etc.). Use a service like OAuth2 or JWT to generate tokens for trusted users. Configure the reverse proxy to validate these tokens before allowing traffic throu.." I tried to install nging and jwt, but i got an error saying .. is it HELD by another process XXXX Is that because comfyUI is running in the back? Anyway, I could not even INSTALL the libraries you recomanded me to install (in another comment), how do I solve that?
Question: Can we create a firewall rule to only give access to a certain "id" or for certain users only? Some "token" that changes all the time, is that possible?
I would have liked if you showed and explained to us (a bit) what the install file does, I appreciate the steps written on the repository, but I need to "learn" more and understand everything going on, if you get me?
I'm considering the n1-highmem-4 or -8 presets, which seem a couple dollars cheaper than the exact same custom ones. Do you find that the number of vCPUs is significant?
Questions again, (do you prefer i send message on linked in or discord?), there are my questions: 1) can I automate login through a third party program, without having to log in to the CLI manually? For example a script running in the pc of my grand mother, she just run the script and she is automatically connectd to the cloud and ready to use comfy. 2) you mentioned possible 30 $ a month right? if its used 3H a day. Ok what if its NOT USED at all, but you sill have that comfy installation on their hard disk (on their server), how much would that cost approx? 3) I asked a follow up question on scaling gpu (under another commnent) Thanks in advance.
Hi NS, I wanted to ask you something PLZ. Today, after starting my virtual machine, I couldn't load ComfyUI through the external IP, even though the VM started normally. Could this be related to the InsightFace installation I just completed?
Probably not, sign in your google cloud and go to compute engine. Check if your machine is started indeed, Check the external id. If all is good You should read the TroubleShooting part in the the repo and do the steps in there
I have tried following this and the instructions on the github multiple times and every time I end up with a set up that seems to be installed and functioning fine through SSH but the web url is unaccessible and times out. I have the port8188 firewall rule set and a static IP set for the VM. The only thing I noticed is that when manually starting Comfy it doesn't know the external IP it just says to use 0.0.0.0:8188. I don't know if that's noteworthy or not. Any help is appreciated!
hi, dm me in telegram( same as my github name) or in linkedin and i will take a look. terminal showing 0.0.0.0:8188 is normal. But it should also show a second one with comfyui external ip.
Hi, I have a question, how do i add/find the jupyter notebook to be able to make custom edits and change the code? BTW, this is the best video explaination for this setup on youtube, thank you so much for making this !
Thank you blitz, What jupyter notebook are you talking about? If you want to edit code files you can SSH into VM and find the path of the file and use nano command to make simple edits.
@@ns_the_one i guess what im asking is, when i have used comfyui on runpod to rent gpus you get a port 3000 that has the code so i can edit directly instead of using the UI elements, its been helpful with some file structures that i need for certain models (Mainly mimic motion),
Google cloud offers a 300$ free trial that doesn't allow you to increase gpu quota. It says that the credits carry over after you upgrade to a paid plan. Do the free trial credits apply to gpus after you upgrade to a paid plan?
@@ns_the_one thanks for your answer! Definitely GC is much more intuitive. And this setup does not stop when the VM is not in use; right? It has to be stopped manually aways to avoid overcharges?
@@Nevalkar_ yes. i think It can be automatized in such way > 1. if you open a tab in your browser for comfyui, first check if VM is started already, if not, start it. 2. send stop the VM command automatically every 10 minutes unless browser has comfyui open.
Hey, man. I've done everything, but I’ve stumbled across a permission issue. In Google Cloud SDK, I connect my Google Cloud account, and when I run gcloud config set project [name-of-my-project], it gives me this error: "WARNING: [my email] does not have permission to access project [my project] (or it may not exist): The caller does not have permission. This command is authenticated as [my email], which is the active account specified by the [core/account] property. Are you sure you wish to set property [core/project] to [my project]?" The thing is, I’m the owner of the Compute Engine project. I’ve also tried using another email, but I get the same result. I double-checked that I have the right permissions, but nothing seems to work. Has this ever happened to you?
Hey amazing tutorial thanks!! I have a question, if i want to download checkpoints from huggingface instead than from civitai, how should i do it? from terminal or from comfyUI interface?
Hi thank you, go to the repo given in description and if you scroll down you will see a section named "How can I download new models". If you have problems feel free to ping me
Thank you for your tutorial. I have installed ComfyUI on Google Cloud, but I want to connect to the TouchDesigner on my local computer and use the camera for real-time generating. I found that the local TouchDesigner can recognize but cannot run my workflow on GoogleCloud ComfyUI. What should I do?
you can probably solve this port forwarding. You need to learn which ports touchdesigner is using to communicate with your camera. Key term is called LocalForward
Another question: You mentioned that you can scale up the gpu and ram etc if needed (if the workflow gets complicated), do you mean we can do that withotu even having to change anythign in the installation? so we dont need to to make a new "server"? If yes that's great. Bonus question to this: is google the least expensive for this? Can we do a similar thing with amazon?
Yes you can. Only thing you need to do is stop your VM and then click edit, pick different hardware and save (You cant change your availability zoe and your availability zone might not all the hardware ) yes you can do this with AWS too.
@@ns_the_one Can you change the VM gpu power with COMMANDS? What I mean is, what if you want your VP capabilities to change depending on certain parameters, so it will detect automacially the need to reduce to increase the VP GPU, instead of having to do that manually?
Thank you for this. I have managed to install a VM on Google Cloud and I am about to install ComfyUI on it. Before I do I wondered if you are almost ready to post your updated video about using this with an API ? I have a client who wants me to set up an API to connect a ComfyUI workflow to a mobile app using a VM on GCP . Any assistance you can offer would be a great help.
@@ns_the_one thank you. I have managed to install ComfyUI and comfyui-on-cloud using your instructions but at the end I received an error here " ./comfyui-on-cloud/src/install.sh: line 106: jq: command not found (23) Failed writing body ---------------- For health check go to this address :8188 in your browser" and I couldn't access Comfy in my browser, do you have a fix for it please ?
gcloud compute scp /Users/username/Documents/my_lora_file ns@comfyvm:/home/ns/ComfyUI/vm_lora_folder running this in your terminal should work if you change the folder and file paths as well as your comfyvm username (ns part ) correctly
@@ns_the_one Yes, this is a new account. And yes, Payment info is added. But the system does not allow me to even give 1 value for GPUs. Enter a new quota value between 0 and 0. Based on your service usage history, you are not eligible for a quota increase at this time. If additional resources are needed, contact our Sales Team to discuss your options for a higher quota value. Created a VM with GPU, and tried applying for quota increase from there. Don't think it worked.
@@ns_the_one Fair enough - thank you for your answer. I tried following the troubleshooting section because I can't access the comfyui via IP address once it's running, but when I go through the steps of "systemctl stop comfyui.service" it asks for a password, which I don't recall setting?
hey when i copy my firewall into my browser i get a "site cannot be reached" error - any workarounds you'd recommend??? i have created the firewall rule for tcp port 8188 (direction = ingress)
I have built a website that deals with ai image generation, I have workflows designed for the specific task it is meant for. But I am running them on local computer. I want to completely automate the service such that without any manual interaction, the whole process will be carried out smoothly. For example, anyone can come on my website, edit with parameters allowed to them, and the whole generation process will take place, and they will get their results. Could you help me with what steps I need to follow to completely achieve this dream
@ns_the_one no, what I meant to say was, my website let's you upload image, and then those image are used as input in comfyui workflow to generate a specific type of image, which is promised to users on my website, and then I deliver the generated images to the users. Now inside the workflow all the values and nodes are fixed, but the initial image to work upon changes from user to user, so such variables(in my case the initial input image) which I want the user to have influence upon I called as parameters, I guess you may have confused it with the parameter inside nodes of workflow, I am sorry for not being able to convey clearly earlier
I closed the terminal after installing everything and had it working, I was trying to install a checkpoint, which I believe I did right, but I needed to restart, so I closed the termainl because it woudnt stop and start via commands like int he video, but now everything is gone, ComfyUI is still working though because I stopped and started via google, how do I get it all back up on terminal?
normally you wont need to do sth extra. gcloud sdk saves your credentials to some folder and you can start the vm with this : "gcloud compute ssh comfyvm". However, if for some reason it is not working, try to do this: "gcloud compute instances list" This outputs list of your compute engines. Check if you see comfyvm there. If you dont see it, it means you need to go through "gcloud auth" or "gcloud init" commands depending on the error you see there.
Hello, i have encountered a problem that i was hoping you maybe could help me with. I had an issue to connect to the ip adress on chrome, It says that i cant connect and i need to check my proxy and firewall. I did it according to the video. I am using windows. Do you maybe have an idea on what the problem could be?
You forgot to make the IP address static and after some time it changed by itself. Check 10:21 in the video. If it is still not working you can book a meeting with me in here: cal.com/ns-the-one
@@ns_the_one thank you for answering! I did it like in the video so it is setup right, but i still cant connect it seems, could there be a firewall issue from my local wifi or firewall issue from my local computer? I live in sweden and i use central europe, could that cause a disturbance?
@@Greatforkman there are a couple of ways to understand the problem. You can book a meeting for free and we can take a look. But If you want to solve it by yourself go to the repo and scroll down and find the troubleshooting section. This section shows debugging steps. And also see the commented lines here : github.com/karaposu/comfyui-on-cloud/blob/main/src/virgin_vm.sh I added them because I once had a similar problem with port access. They might help. I paste them here: ## run these lines in the vm to make sure port 8188 is open for external access #sudo ufw allow 8188/tcp #sudo ufw reload #sudo firewall-cmd --zone=public --add-port=8188/tcp --permanent #sudo firewall-cmd --reload #sudo apt-get install iptables #sudo iptables -A INPUT -p tcp --dport 8188 -j ACCEPT
@@ns_the_one Thank you for the help! I found different reasons as to what could have been the problem. First thing was that SSH had was not installed properly as since i got connection issues randomly when trying to connect to VM. The next was that 100gb space on the compute engine was not enough to download everything (I got an error message), so i had to raise disc space to 120gb. So i had experimented a bit, i thought it would be safest to start over. So i removed the project and everything and started from scratch. This time with the 120gb extra disc space and better understanding how to setup the SSH. And with doing all of this, i managed to get everything working at the end of it. Great guide and thank you for the help! Next thing im trying to figure out and what you could a video about is how to make VM comfyui work with Krita.
@@Greatforkman Yeah, after some updates 100gb is not enough anymore. I am glad it worked. Please star the repo in github if you have an account and have fun generations!
Absolutely, I am creating a new tutorial specifically about this. For starters you can check: github.com/pydn/ComfyUI-to-Python-Extension In new tutorial i will show how to turn your comfyui workflow to a production-grade backend API with multiuser support
@ns_the_one can't wait for that ! And thanks for this tutorial! Well ,I have simply developed a Gradio app which connect to my local Comfyui (as a API call). While I'm looking for a cloud hosting service and I found this video, it really helpful (I have almost no coding experience). Appreciate for that!!!
@@euroronaldauyeung8625 feel free to connect through linkedin. I am also launching my SD based app soon and would be cool to connect with people with similar goals.
@@ns_the_oneHi Bro, I have installed comfyui in GCP, it can run properly! However, I found that it keep disconnect , especially it have been idle for 10-15 min or after a big workflow is run.... do you experienced the same issue?
@@euroronaldauyeung8625 nope, mine is open for days and running flows time 2 time and i have no disconnections. How big is your workflow? It might be that you need to increase the RAM in for your compute instance
PSA for those using this method and are wanting to use Stable Cascade or SDXL: I would highly recommend setting your instance RAM to 24GB (and have 8 vcores) as this will keep your cloud server from crashing.
Hocam ben Comfyui'da proje yaptım, onu cloud ile website yapıp eklemek istiyorum, aylık üyelik ile o projeyi kullandırmak istiyorum. Buda bu yöntem ile mi yapılıyor? Pek bir kaynak bulamadım bu konu hakkında.
Merhaba dostum, bende tam onunla alakali bir tutorial hazirlama asamasindayim. Kisaca yapman gereken sey su, bir backend server da comfyui calisiyor olucak ve input output lar API vasitasi ile verilicek. Yapilis yonteminde bazi temel farklar var. Detayli konusuruz istersen
@@sassss5984 GPU ile generation yapacaksan illaki masrafli olucak. Ama eger MVP cikariyorsan 1 GPU yeterli olur en basta. Kullanim miktarina gore yeni GPU lar eklersin zamanla. Videoyu begenip ve github linkini star larsan sevinirim dostum.
something that's more interesting to me (and possibly to a lot of people) is deploying ComfyUi as API endpoint on GCP. Do you know how we can do that, so we run our workflows as API endpoints in the cloud? This way we can create SaaS apps with GCP hosted comfyui backend
@@pikushaa you can remove some models to open up space. But if you are going to explore new models i suggest redoing the installation with bigger disk space.
@@ns_the_one Hey man, really appreciate the work you put in to bring this to others. I'm finally able to learn and play around with comfyui without having to use colab. I have a question, how do I remove models?
Was so happy to find this video until i was notified free trial accounts have limited quota and cannot increase. Can still try with their TPU with 128GB memory tho still better than my macbook pro intel lol
are lost during installation of gcloud? it should be straight forward. And once you install it, the rest is 4 lines of code. Write to me the exact step you got lost and i will help you find the track
Hi! Thanks for the great tutorial, it is really game changing. Also I found a way to connect VM via ftp manager, maybe that will be helpful: ua-cam.com/video/LzFRFdgg8Wo/v-deo.html I have one question, Is there any way to see a Comfyui terminal where it working? There's a lot of important information there. Thanks again :)
Hey folks,
I’m planning to update the installation files with the latest features. It would be great if you could suggest some of the latest, reliable extensions and models. I was considering including FLUX, but I think it might be too large and costly for most beginners. So, I’m thinking of providing two versions of the installation files: a standard minimal version (16GB RAM, 120GB storage) and a more advanced version (32GB RAM, 200GB storage).
Let me know your thoughts!
- Flux shcnell is less large then flux dev no?
- Can you keep the simple one and make another version? I like to have simple things to learn, but maybe it is best to upgrade? your call.
- I would like to learn how to "control" who gets access. I dont want any person with the ip to be able to enter, but I dont want also to have only one ip to be able. There must be something else that can say if someone can enter or not, I dont knwo yet.
- Tell me, in that CLI black window I could theoreticaly create a python script of my choice that can itself control the comfy right?
- question what if you want to gather mulriple prompt thourgh the day, and only want the server to run them all at once at the end of the day, would you PAY for the server beign run despite not using the GPU through the day (just gathering prompts)?
- sorry for many questions, I am excited about this topic
@@EternalAI-v9b Yup, it is configurable but most of the ways require some extra steps in terminal so I avoid sharing them. There is 1 feasible-looking way i want to check:
"API Gateway or Reverse Proxy with Token Authentication:
If you're exposing certain services via HTTP/HTTPS (e.g., a web server on the VM), you can front those services with an API gateway or reverse proxy that enforces authentication via dynamically generated tokens or OAuth.
Steps:
Set up an API gateway (such as Kong, NGINX with JWT, etc.).
Use a service like OAuth2 or JWT to generate tokens for trusted users.
Configure the reverse proxy to validate these tokens before allowing traffic through."
Through this we might create a password and only the ones with password can access.
@@ns_the_one Impressive! I love it^^ I will copy paste it. I dont know well all this but I am sure I will be able to understand it in the future.
I would definitely include FLUX dev, I see it being used a lot for photorealism.
Also, it would be great to let us know what happens if we update and change the GPU to add some faster ones and increase the memory which would lead to changing the location.
For the most part, you can't do much with basic libraries and at some point, you have to agree that you have to spend more if you want to achieve good results.
@ns_the_one Flux has GGUF and many other versions that you can use for a tutorial.
something that's more interesting to me (and possibly to a lot of people) is deploying ComfyUi as API endpoint on GCP. Do you know how we can do that, so we run our workflows as API endpoints in the cloud? This way we can create SaaS apps with GCP hosted comfyui backend
Using this video as inspiration, I set up a g2-standard-12 instance with the L4 processor; it's 4x the speed of the T4. Costs more per hour, but still less than $1, and with 24GB of VRAM, it's capable of running the full hunyuan video model, and perhaps even Nvidia Cosmos. Note that you can save a few cents and increase security by not reserving a static IP address. Instead, create a loopback by adding --ssh-flag="-L 8188:localhost:8188" to the "gcloud compute ssh" command, and then just connect to localhost:8188 on your local machine.
Oh my gosh this actually worked. Not gonna lie, I nearly gave up when you got to the Terminal part because I know absolutely nothing about code and have only used Terminal once or twice because I find it so intimidating but I pushed through and now it's up and running on my intel mac! Didn't think it was possible to use ComfyUI, or any stable diffusion on this computer - thank you!
I am glad it worked. I intentionally structured the tutorial for non-developer people. Have fun image generations!
Thanks a lot for this. Looking forward to your tutorial on production grade API!
thx for the support mate. It is still in progress
@@ns_the_one hello! curious, is the production grade api tutorial out yet?
wow very nice... you're the superman bro.
Thank you!
I read in comments you would be making another video for comfyUI with GCP as API backend... can't wait to see it.
You are a legend man, you just saved me spending thousands on a new macbook.
you are welcome mate
@@ns_the_one You are a legend because I just searched "install comfy ui on your own sever" and was afraid because I didnot finf any good result within the 3 first results, thankfully I found yours. I will watch it now. What I liked the most is the sentece you wrote on your video descripttion about its cost. I really like that. I did not watch the video fully yet, but can we customize the server and add our own pyton scripts etc? I wonder. Will watch and see
@@EternalAI-v9b haha thanks, i just youtubed "install comfy ui on your own sever" couldnt find my video as well. Yes you can customize it. This tutorial, in it's essence, is showing you how to create a virtual machine on cloud -with GPU support- and use it for random stuff.
Hey guys, there was a small error in the code which i thought i fixed it but it wasnt fixed. Well, this time it is fixed for sure! Everything should be working well without any issues by now. Still, please leave a comment if you encounter any problems.
thank you so much !!! youre a legend
I can not increase the quota, even after requesting and paying $10. Is still gonna work without increasing it?
@@entertainmentchannel9632 You mean GPU quota? I can suggest to change the text in the explanation field. Write there some unique text about experimenting with GPUs. Also which country are you accessing the GCP from ?
@@ns_the_one UK
@@entertainmentchannel9632 cal.com/ns-the-one
and lets check it together
You rock bro ! This tutorial was very helpful.
you are amazing, thank you for this !!!
Man, this is epik. Thank you so much, the video tutorial was very much needed
Absolutely fantastic tutorial and resource. Precisely what I needed thank you so much!
Thank you, give it a star in github if it was helpful
If anyone has to stop to wait for the GPU quota request to be accepted, here is where the video continues: 8:37
Çok net ve başka yerde bulunamayacak bir içerik, eline sağlık.
Eyvallah dostum dusunmen yeter. Linkedin'den ekleyebilirsin, bir app cikariyorum yakinda denersin yorum yaparsin
@@ns_the_one istek gönderdim, tabii ki denerim
Thanks for the speed comparison!
Thanks for creating this💪. I just made it work but total installation took around 1 hour. And as far as I understand, if you create a new GCP account, storage and CPU cost will consume your free credits and wont cost you extra.
Question about authentificaiton (13:19), I read somewhere that if you create multiple google account and multiple google cloud accounts with the same credit card, gogole mmight terminate the account becuase you are trying to get the free tier in duplicate perhaps. In this case I see you using 2 accounts, did you choose the one where you had a google cloud account, and left the other account where you dont have any DEV account or whatsoever? Curious. Maybe I am wrong
This is amazing, thank you! Do you think that also the rendering of AI Videos could be suffraged onto a cloud server? I would be super curious about a material about it!
yup, you need a choose a higher RAM GPU though. And tricky part is not all GCP regions allow selecting all types of GPUs.
fantastic! thank you for this.
if I wanted to install Automatic1111 & Fooocus alongside your setup, anything I should keep in mind as to not break anything and be able to import/use the same models ?
Wow thanks! This is what I was looking for! How can I access the output images folder? and is there a workflow to upload them to a service like Google Drive to download them on my Mac?
Thx for the reminding me. I just updated the readme file accordingly. Here you go :
1. Find out the full path of your ComfyUI output folder. Mine is `/home/ns/ComfyUI/output`.
2. Create a new folder in your local machine and cd into that directory through terminal
3. run this command (change "ns" with your comfyvm username and also change the output directory path accordingly )
```gcloud compute scp --recurse ns@comfyvm:/home/ns/ComfyUI/output ./```
This will download the images into your local machine.
@@ns_the_one - will it also delete them? Or due you have another terminal command to --recurse, then delete. I mainly want to do large batches and upscales, to be refined in Photoshop, so I see no reason to keep them on GDrive after downloading.
BTW: great video and perfect timing for me. Moving to the next stage of AI-gen and a. Mac is the worst for this, but best for everything else; b. I don't want to maintain a 'Doze PC/4090 (or even Linux) just for AI. CloudGPUs are the way to go.
EDIT: ChatGPT just gave me this code to delete if anyone is wondering:
"gdrive list | grep | awk '{print $1}' | xargs -I {} gdrive delete {}"
@@TheDocPixel The code you show is about gdrive. The code i provided is to download images to your local folder. Once you download images you can simply run
gcloud compute ssh comfyvm --command "rm -rf ComfyUI/output/*"
to delete the images inside the output folder of comfyui
11:33, you choose 8188, because of comfyUI uses this port in the browser, or does that have nothing to do with it? Sorry I am kind of learning.
yup thats default port for comfyui
HI, you told me about these steps:
"Set up an API gateway (such as Kong, NGINX with JWT, etc.).
Use a service like OAuth2 or JWT to generate tokens for trusted users.
Configure the reverse proxy to validate these tokens before allowing traffic throu.."
I tried to install nging and jwt, but i got an error saying .. is it HELD by another process XXXX
Is that because comfyUI is running in the back?
Anyway, I could not even INSTALL the libraries you recomanded me to install (in another comment), how do I solve that?
Hi friend, get an the error. I can't edit quota.
Pure gold
Appreciate it
Question: Can we create a firewall rule to only give access to a certain "id" or for certain users only? Some "token" that changes all the time, is that possible?
I would have liked if you showed and explained to us (a bit) what the install file does, I appreciate the steps written on the repository, but I need to "learn" more and understand everything going on, if you get me?
Hey this works but each time I close/stop my vm and reboot I have to reinstall comfyui and models. Not sure what I’m doing wrong
You shouldnt have to do that. DM me from telegram or linkedin and I will check it out.
I'm considering the n1-highmem-4 or -8 presets, which seem a couple dollars cheaper than the exact same custom ones. Do you find that the number of vCPUs is significant?
I honestly dont know. But I would guess no big impact on the performance.
when im in the terminal and installing all comfy stuff i get disconnected after like 2mins, everytime...
and we can use API call for the web application
Questions again, (do you prefer i send message on linked in or discord?), there are my questions:
1) can I automate login through a third party program, without having to log in to the CLI manually? For example a script running in the pc of my grand mother, she just run the script and she is automatically connectd to the cloud and ready to use comfy.
2) you mentioned possible 30 $ a month right? if its used 3H a day. Ok what if its NOT USED at all, but you sill have that comfy installation on their hard disk (on their server), how much would that cost approx?
3) I asked a follow up question on scaling gpu (under another commnent)
Thanks in advance.
+1 for these questions
Hi NS, I wanted to ask you something PLZ. Today, after starting my virtual machine, I couldn't load ComfyUI through the external IP, even though the VM started normally. Could this be related to the InsightFace installation I just completed?
Probably not, sign in your google cloud and go to compute engine. Check if your machine is started indeed, Check the external id. If all is good You should read the TroubleShooting part in the the repo and do the steps in there
hello sir I want to make web application on the comfyui so same type of cloud server ok
I have tried following this and the instructions on the github multiple times and every time I end up with a set up that seems to be installed and functioning fine through SSH but the web url is unaccessible and times out. I have the port8188 firewall rule set and a static IP set for the VM. The only thing I noticed is that when manually starting Comfy it doesn't know the external IP it just says to use 0.0.0.0:8188. I don't know if that's noteworthy or not.
Any help is appreciated!
hi, dm me in telegram( same as my github name) or in linkedin and i will take a look. terminal showing 0.0.0.0:8188 is normal. But it should also show a second one with comfyui external ip.
I got it to work, but everything has to be followed exactly. The VM name has to be the same and the zone has to be europe central or it breaks.
Thanks so much, finaly ai-ing on an i3 laptop :D
Bạn đến từ Việt Nam à?
Hi, I have a question, how do i add/find the jupyter notebook to be able to make custom edits and change the code? BTW, this is the best video explaination for this setup on youtube, thank you so much for making this !
Thank you blitz, What jupyter notebook are you talking about? If you want to edit code files you can SSH into VM and find the path of the file and use nano command to make simple edits.
@@ns_the_one i guess what im asking is, when i have used comfyui on runpod to rent gpus you get a port 3000 that has the code so i can edit directly instead of using the UI elements, its been helpful with some file structures that i need for certain models (Mainly mimic motion),
Google cloud offers a 300$ free trial that doesn't allow you to increase gpu quota. It says that the credits carry over after you upgrade to a paid plan. Do the free trial credits apply to gpus after you upgrade to a paid plan?
Hi, amazing tutorial! Just one curiosity, why gc instead of azure or aws? Have you compare pricing?
imo GCP interface is much more intuitive compared to AWS. Idk about pricing tho
@@ns_the_one thanks for your answer! Definitely GC is much more intuitive. And this setup does not stop when the VM is not in use; right? It has to be stopped manually aways to avoid overcharges?
@@Nevalkar_ yes. i think It can be automatized in such way >
1. if you open a tab in your browser for comfyui, first check if VM is started already, if not, start it.
2. send stop the VM command automatically every 10 minutes unless browser has comfyui open.
Hey, man. I've done everything, but I’ve stumbled across a permission issue. In Google Cloud SDK, I connect my Google Cloud account, and when I run gcloud config set project [name-of-my-project], it gives me this error:
"WARNING: [my email] does not have permission to access project [my project] (or it may not exist): The caller does not have permission. This command is authenticated as [my email], which is the active account specified by the [core/account] property. Are you sure you wish to set property [core/project] to [my project]?"
The thing is, I’m the owner of the Compute Engine project. I’ve also tried using another email, but I get the same result. I double-checked that I have the right permissions, but nothing seems to work.
Has this ever happened to you?
message me via telegram or linkedin and lets take a look at it. My telegram nickname is same as my github name
Hey amazing tutorial thanks!! I have a question, if i want to download checkpoints from huggingface instead than from civitai, how should i do it? from terminal or from comfyUI interface?
Hi thank you, go to the repo given in description and if you scroll down you will see a section named "How can I download new models". If you have problems feel free to ping me
Thank you for your tutorial. I have installed ComfyUI on Google Cloud, but I want to connect to the TouchDesigner on my local computer and use the camera for real-time
generating. I found that the local TouchDesigner can recognize but cannot run my workflow on GoogleCloud ComfyUI. What should I do?
you can probably solve this port forwarding. You need to learn which ports touchdesigner is using to communicate with your camera. Key term is called LocalForward
Another question: You mentioned that you can scale up the gpu and ram etc if needed (if the workflow gets complicated), do you mean we can do that withotu even having to change anythign in the installation? so we dont need to to make a new "server"? If yes that's great.
Bonus question to this: is google the least expensive for this? Can we do a similar thing with amazon?
Yes you can. Only thing you need to do is stop your VM and then click edit, pick different hardware and save (You cant change your availability zoe and your availability zone might not all the hardware )
yes you can do this with AWS too.
@@ns_the_one Can you change the VM gpu power with COMMANDS? What I mean is, what if you want your VP capabilities to change depending on certain parameters, so it will detect automacially the need to reduce to increase the VP GPU, instead of having to do that manually?
@@EternalAI-v9b not possible in a feasible way
@@ns_the_one What's the alternative?
Thank you for this. I have managed to install a VM on Google Cloud and I am about to install ComfyUI on it. Before I do I wondered if you are almost ready to post your updated video about using this with an API ? I have a client who wants me to set up an API to connect a ComfyUI workflow to a mobile app using a VM on GCP . Any assistance you can offer would be a great help.
Code base for multiuser api is partially ready. But I am working on another project for now.
@@ns_the_one thank you for your quick update, I shall look forward to it when it's ready. Cheers !
@@ns_the_one thank you. I have managed to install ComfyUI and comfyui-on-cloud using your instructions but at the end I received an error here " ./comfyui-on-cloud/src/install.sh: line 106: jq: command not found
(23) Failed writing body
---------------- For health check go to this address :8188 in your browser" and I couldn't access Comfy in my browser, do you have a fix for it please ?
Hello,I want to download a Lora that I trained myself, but how do I download it into my comfy UI?
files are in your local?
gcloud compute scp /Users/username/Documents/my_lora_file ns@comfyvm:/home/ns/ComfyUI/vm_lora_folder
running this in your terminal should work if you change the folder and file paths as well as your comfyvm username (ns part ) correctly
@@ns_the_one Thank you for your helpful response, but I have one more question: how can I locate the LoRA folder within ComfUI?"
Based on your service usage history, you are not eligible for a quota increase at this time.
Ideas on how to fix this ?
is it a new account? You entered your correct payment information already? What text you wrote for your GPU quota change request reason?
@@ns_the_one Yes, this is a new account. And yes, Payment info is added. But the system does not allow me to even give 1 value for GPUs.
Enter a new quota value between 0 and 0. Based on your service usage history, you are not eligible for a quota increase at this time. If additional resources are needed, contact our Sales Team to discuss your options for a higher quota value.
Created a VM with GPU, and tried applying for quota increase from there. Don't think it worked.
Works great! Thank you! Is there any way to prevent access to the external ip? As far as I understand, everyone with the ip can use it correct?
Glad it worked, easiest solution is to add your static ip address to the part i show in 11:18
Is there a reason you picked GCP over lambda? just curious about the potential difference
GCP VMs are easier to use/manage imo
@@ns_the_one Fair enough - thank you for your answer. I tried following the troubleshooting section because I can't access the comfyui via IP address once it's running, but when I go through the steps of "systemctl stop comfyui.service" it asks for a password, which I don't recall setting?
hey when i copy my firewall into my browser i get a "site cannot be reached" error - any workarounds you'd recommend??? i have created the firewall rule for tcp port 8188 (direction = ingress)
Yeah, it was a new bug introduced after fixing another small bug. I just pushed a new commit and it is fixed right now.
I have built a website that deals with ai image generation, I have workflows designed for the specific task it is meant for. But I am running them on local computer. I want to completely automate the service such that without any manual interaction, the whole process will be carried out smoothly. For example, anyone can come on my website, edit with parameters allowed to them, and the whole generation process will take place, and they will get their results. Could you help me with what steps I need to follow to completely achieve this dream
when you say parameters, are they workflow parameters used in comfyui?
@ns_the_one no, what I meant to say was, my website let's you upload image, and then those image are used as input in comfyui workflow to generate a specific type of image, which is promised to users on my website, and then I deliver the generated images to the users. Now inside the workflow all the values and nodes are fixed, but the initial image to work upon changes from user to user, so such variables(in my case the initial input image) which I want the user to have influence upon I called as parameters, I guess you may have confused it with the parameter inside nodes of workflow, I am sorry for not being able to convey clearly earlier
@@rambapat588 you can wait for my next tutorial which will be about this exactly. But I am not sure when I will release it
I closed the terminal after installing everything and had it working, I was trying to install a checkpoint, which I believe I did right, but I needed to restart, so I closed the termainl because it woudnt stop and start via commands like int he video, but now everything is gone, ComfyUI is still working though because I stopped and started via google, how do I get it all back up on terminal?
Its ok, managed to fix it via the docs
normally you wont need to do sth extra. gcloud sdk saves your credentials to some folder and you can start the vm with this : "gcloud compute ssh comfyvm". However, if for some reason it is not working, try to do this:
"gcloud compute instances list" This outputs list of your compute engines. Check if you see comfyvm there.
If you dont see it, it means you need to go through "gcloud auth" or "gcloud init" commands depending on the error you see there.
Hello, i have encountered a problem that i was hoping you maybe could help me with. I had an issue to connect to the ip adress on chrome, It says that i cant connect and i need to check my proxy and firewall. I did it according to the video. I am using windows. Do you maybe have an idea on what the problem could be?
You forgot to make the IP address static and after some time it changed by itself. Check 10:21 in the video. If it is still not working you can book a meeting with me in here: cal.com/ns-the-one
@@ns_the_one thank you for answering!
I did it like in the video so it is setup right, but i still cant connect it seems, could there be a firewall issue from my local wifi or firewall issue from my local computer?
I live in sweden and i use central europe, could that cause a disturbance?
@@Greatforkman there are a couple of ways to understand the problem. You can book a meeting for free and we can take a look. But If you want to solve it by yourself go to the repo and scroll down and find the troubleshooting section. This section shows debugging steps.
And also see the commented lines here : github.com/karaposu/comfyui-on-cloud/blob/main/src/virgin_vm.sh
I added them because I once had a similar problem with port access. They might help. I paste them here:
## run these lines in the vm to make sure port 8188 is open for external access
#sudo ufw allow 8188/tcp
#sudo ufw reload
#sudo firewall-cmd --zone=public --add-port=8188/tcp --permanent
#sudo firewall-cmd --reload
#sudo apt-get install iptables
#sudo iptables -A INPUT -p tcp --dport 8188 -j ACCEPT
@@ns_the_one Thank you for the help!
I found different reasons as to what could have been the problem. First thing was that SSH had was not installed properly as since i got connection issues randomly when trying to connect to VM. The next was that 100gb space on the compute engine was not enough to download everything (I got an error message), so i had to raise disc space to 120gb.
So i had experimented a bit, i thought it would be safest to start over. So i removed the project and everything and started from scratch. This time with the 120gb extra disc space and better understanding how to setup the SSH.
And with doing all of this, i managed to get everything working at the end of it.
Great guide and thank you for the help!
Next thing im trying to figure out and what you could a video about is how to make VM comfyui work with Krita.
@@Greatforkman Yeah, after some updates 100gb is not enough anymore. I am glad it worked. Please star the repo in github if you have an account and have fun generations!
Thank you will save this video
This is awesome. Any way to DM you outside of LinkedIn ?
Thank you, you can reach through reddit or telegram too. my username for both is karaposu
Is that possible to deploy the comfyui in GCP in this way which act as a API backend?
Absolutely, I am creating a new tutorial specifically about this. For starters you can check: github.com/pydn/ComfyUI-to-Python-Extension
In new tutorial i will show how to turn your comfyui workflow to a production-grade backend API with multiuser support
@ns_the_one can't wait for that ! And thanks for this tutorial! Well ,I have simply developed a Gradio app which connect to my local Comfyui (as a API call). While I'm looking for a cloud hosting service and I found this video, it really helpful (I have almost no coding experience). Appreciate for that!!!
@@euroronaldauyeung8625 feel free to connect through linkedin. I am also launching my SD based app soon and would be cool to connect with people with similar goals.
@@ns_the_oneHi Bro, I have installed comfyui in GCP, it can run properly! However, I found that it keep disconnect , especially it have been idle for 10-15 min or after a big workflow is run.... do you experienced the same issue?
@@euroronaldauyeung8625 nope, mine is open for days and running flows time 2 time and i have no disconnections. How big is your workflow? It might be that you need to increase the RAM in for your compute instance
PSA for those using this method and are wanting to use Stable Cascade or SDXL: I would highly recommend setting your instance RAM to 24GB (and have 8 vcores) as this will keep your cloud server from crashing.
Hocam ben Comfyui'da proje yaptım, onu cloud ile website yapıp eklemek istiyorum, aylık üyelik ile o projeyi kullandırmak istiyorum. Buda bu yöntem ile mi yapılıyor? Pek bir kaynak bulamadım bu konu hakkında.
Merhaba dostum, bende tam onunla alakali bir tutorial hazirlama asamasindayim. Kisaca yapman gereken sey su, bir backend server da comfyui calisiyor olucak ve input output lar API vasitasi ile verilicek. Yapilis yonteminde bazi temel farklar var. Detayli konusuruz istersen
@@ns_the_one Tutorial videosunu bekliyorum o zaman. Çok masraflı olmadan bu yapılabilinirse süper olur.
@@sassss5984 GPU ile generation yapacaksan illaki masrafli olucak. Ama eger MVP cikariyorsan 1 GPU yeterli olur en basta. Kullanim miktarina gore yeni GPU lar eklersin zamanla. Videoyu begenip ve github linkini star larsan sevinirim dostum.
Kubernetes ile scalable da yapılabilir sanırım
something that's more interesting to me (and possibly to a lot of people) is deploying ComfyUi as API endpoint on GCP. Do you know how we can do that, so we run our workflows as API endpoints in the cloud? This way we can create SaaS apps with GCP hosted comfyui backend
Google wont let me increase the GPU quota
what is the error
how to I install models?
explained here : github.com/karaposu/comfyui-on-cloud
@@ns_the_one [Errno 28] No space left on device, please help to fix this
@@pikushaa you can remove some models to open up space. But if you are going to explore new models i suggest redoing the installation with bigger disk space.
@@ns_the_one Hey man, really appreciate the work you put in to bring this to others. I'm finally able to learn and play around with comfyui without having to use colab.
I have a question, how do I remove models?
@@thalla1asal1wolf i am glad it is useful. Login to your server and run this
"cd ComfyUI"
and then run
"rm -rf ./models/checkpoints/model_name"
Was so happy to find this video until i was notified free trial accounts have limited quota and cannot increase. Can still try with their TPU with 128GB memory tho still better than my macbook pro intel lol
The idea is not, not to pay, but pay the least amount.
I am on Windows, everything is easy to follow up to the terminal section, then I get lost
are lost during installation of gcloud? it should be straight forward. And once you install it, the rest is 4 lines of code. Write to me the exact step you got lost and i will help you find the track
Should I use cmd or powershell? @@ns_the_one
cool
Hi!
Thanks for the great tutorial, it is really game changing.
Also I found a way to connect VM via ftp manager, maybe that will be helpful: ua-cam.com/video/LzFRFdgg8Wo/v-deo.html
I have one question, Is there any way to see a Comfyui terminal where it working? There's a lot of important information there.
Thanks again :)
I came back here to ask the same question
hello, either terminal or git bach cannot read wget command even I have installed google cloud cli. please help
can you paste here command you are running and the error you get?