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.
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.
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 ?
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.
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?
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
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.
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?
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.
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),
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
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
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?
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 $1}' | xargs -I {} gdrive delete {}"
@@TheDocPixel The code you show is about gdrive. The code i provided is to download images to your local folder. Once you download images you can simply run gcloud compute ssh comfyvm --command "rm -rf ComfyUI/output/*" to delete the images inside the output folder of comfyui
Google cloud offers a 300$ 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?
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
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 ?
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
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 ?
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)
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.
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!
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
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
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: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-LzFRFdgg8Wo.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 :)
@@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?