Did something change recently? I was using this installation process with no problems for the past month or so, but in the last few days I can't get it to work. Can't get ComfyUI to open. When I set up the Vast template and start the instance, the "Connecting/Open" button immediately functions as a link to open Jupyter Labs. In the past this button would just say "Connecting" and doesn't work until I go through the full installation, and then it would switch to "Open" and clicking it would open ComfyUI. I try to directly navigate to port 3000 but it redirects to Jupyter Labs again. I've tried restarting Comfy, rebooting the instance, reinstalling everything, but nothing seems to work. I've tested this out on a dozen different instances, both with the manually set up template and the prebuilt Vast template. Anyone else experiencing the same problem?
That's a shame to hear. I'll try and see if I can get time soon to test it myself too. Be great to hear if anyone else has the issue or finds the solution.
If anyone else is having this issue, I figured out the fix. @AIAnimation You'll have to manually edit the image template. Change the Docker options in the template to -e JUPYTER_DIR=/ -p 3000:3000 -e OPEN_BUTTON_PORT=3000
@@AIAnimationStudio I figured it out. You'll have to manually edit the image template. Change the Docker options in the template to -e JUPYTER_DIR=/ -p 3000:3000 -e OPEN_BUTTON_PORT=3000
You say somewhere that you like to install by script rather than the image on vast. I can definitely see some advantages to that, but there are some good ones for image. Installing it doesnt count to your time or storage costs. Ive been using the krita plugin one a lot. Its 27gig comes with multiple models (digital, photo. lcm) and takes ages to install! But that time doesnt cost me anything and the storage space I reserve can be less because I already have two good models. Got to learn myself how all that works.
Where do you see that using an image does not cost your runtime it storage time? I went through the vast ai website and their FAQs and do not see what you described anywhere.
Great video Thank you ! But unfortunately ComfyUI doesn't support yet multi-GPU, then no needs to buy 8x GPU (As comfy is going to use only one of them). Maybe something to do with SwarmUI from stability who is supposed to manage multiGPU ? But your process is still very usefull with one GPU. Fantastic work !
Ah, that's annoying (but good to know). Should have checked that before making the vid. Curerntly running pretty quickly locally with that PyTorch Lora (on a Mac). But will do some speed tests on Vast.ai too for comparison.
Good Job Man! What a great, informative and fast tutorial. The only question I am left with is: How fast are the generation with these 8xRTX 4090. On 06:03 you want to know how fast it is - I want to know that, too :D Please...!
I really do need to do some comaparison of the 2 instances and share the results. Should have for the video but time was getting on. I'll sort that out soon.
Awesome tutorial. I am very grateful. I am wondering: now that I have this up and running, how would I install checkpoints, loras, models that I find on Civitai, etc? I wish to use SDXL in this case, and am not sure where/how to load it. I've attempted to download the code from the huggingface link to the checkpoints2 folder, but running comfy it appears I've made an error. Any help would be deeply appreciated. Thank you!
Thank you for this tutorial! Can I just install Comfy + Manager, then paste my workflow and download the missing custom nodes and models trough the manager?
I found that once I opened up ComfyUI, I was able to 'Load' one of my saved workflows, and use the manager to install any missing custom nodes. If you need help transferring other files like LoRAs or clip_vision files that you can't just get from the manager, I made an extensive reply in this comment chain detailing a process that worked for me for doing this.
Amazing video man.... but do I need to be in the root directory to install and run comfy UI, as I am using sagemaker studio lab and I dont have permission to root directory. Is there no other way to install it without accessing the root directory.
Yeah I believe so. At leat one with ComfUI manager would be helpful, so you cna then use that to install any missing dependencies depending on your loaded json workflow.
Is this only intended to work as an administrator user @aianimation? I installed it under the admin account (with SUDO) but when a standard user tries to use it, only the background comes up with none of the nodes
heyy!! just a problem. Everything was working till one week ago, now when I try to open the gui port, it doesn't open the comfyui interface but the jupyter notebook again, some advice?
Unfortunately not. It's saved to that instance. They're working on ways to have possible to move to other instances (not sure how that will work though). So at the moment if you change instances you have to load everything which only takes around 10 mins. (But if you stop and start the same instance (*assuming you can) the files will be there for you.
@@AIAnimationStudio Loading everything takes 10mins? I'm currently installing the setup and some files are taking over an hour to download! I imagine it's going to take me 5 or 6 hours to get the whole thing set up. Can you explain what you mean by it only taking 10 mins - is it the first time that takes ages but changing instances (I assume that means different GPU configs) only takes 10 mins?
@@Richie_W … no, it shouldn’t take that long by to download the models or anything. From starting from scratch to up and running should be around 10-15 mins . Perhaps the internet speed on the system your renting is crazy slow?? Hmm. I’ll post a new video soon with some other options … if/when I get time.
Is that after going through all the steps? i.e. can you open Jupyter? With Comfy setup and running the last bit of code, should load up Comfy UI and make it available via the connecting/open button.
@@AIAnimationStudio Its all working great now. I think the instance I picked just had network problems, another one worked great. Thanks heaps for the tut. Once I understood whats going on its not to expensive for me (and Im poor) so I can have a play now. It failed on first run, but I figured it just didnt down a model right and regot it, your script is so useful. Thanks!
Did something change recently? I was using this installation process with no problems for the past month or so, but in the last few days I can't get it to work. Can't get ComfyUI to open. When I set up the Vast template and start the instance, the "Connecting/Open" button immediately functions as a link to open Jupyter Labs. In the past this button would just say "Connecting" and doesn't work until I go through the full installation, and then it would switch to "Open" and clicking it would open ComfyUI. I try to directly navigate to port 3000 but it redirects to Jupyter Labs again. I've tried restarting Comfy, rebooting the instance, reinstalling everything, but nothing seems to work. I've tested this out on a dozen different instances, both with the manually set up template and the prebuilt Vast template. Anyone else experiencing the same problem?
That's a shame to hear. I'll try and see if I can get time soon to test it myself too. Be great to hear if anyone else has the issue or finds the solution.
If anyone else is having this issue, I figured out the fix. @AIAnimation You'll have to manually edit the image template. Change the Docker options in the template to -e JUPYTER_DIR=/ -p 3000:3000 -e OPEN_BUTTON_PORT=3000
@@AIAnimationStudio I figured it out. You'll have to manually edit the image template. Change the Docker options in the template to -e JUPYTER_DIR=/ -p 3000:3000 -e OPEN_BUTTON_PORT=3000
@@sportsupyours . Brilliant thanks for sharing.
@@sportsupyours When I make this change in the template I can't get past the instance creation stage, it hangs on "connecting." Thoughts?
brilliant guide thank you! Are all details up to date okay for making a fresh build today?
Awesome video! Is it possible to set up an API request to trigger a workflow and return an image?
You say somewhere that you like to install by script rather than the image on vast. I can definitely see some advantages to that, but there are some good ones for image. Installing it doesnt count to your time or storage costs. Ive been using the krita plugin one a lot. Its 27gig comes with multiple models (digital, photo. lcm) and takes ages to install! But that time doesnt cost me anything and the storage space I reserve can be less because I already have two good models. Got to learn myself how all that works.
Where do you see that using an image does not cost your runtime it storage time? I went through the vast ai website and their FAQs and do not see what you described anywhere.
Great video Thank you ! But unfortunately ComfyUI doesn't support yet multi-GPU, then no needs to buy 8x GPU (As comfy is going to use only one of them). Maybe something to do with SwarmUI from stability who is supposed to manage multiGPU ? But your process is still very usefull with one GPU. Fantastic work !
Ah, that's annoying (but good to know). Should have checked that before making the vid. Curerntly running pretty quickly locally with that PyTorch Lora (on a Mac). But will do some speed tests on Vast.ai too for comparison.
Good Job Man! What a great, informative and fast tutorial. The only question I am left with is: How fast are the generation with these 8xRTX 4090. On 06:03 you want to know how fast it is - I want to know that, too :D Please...!
I really do need to do some comaparison of the 2 instances and share the results. Should have for the video but time was getting on. I'll sort that out soon.
Awesome tutorial. I am very grateful. I am wondering: now that I have this up and running, how would I install checkpoints, loras, models that I find on Civitai, etc? I wish to use SDXL in this case, and am not sure where/how to load it. I've attempted to download the code from the huggingface link to the checkpoints2 folder, but running comfy it appears I've made an error. Any help would be deeply appreciated. Thank you!
Hey, tricky to help without seeing your exact setup. Was Comfy running with the other checkpoints? before trying SDXL?
But comfy ui is not multi gpu. How does it works when renting more than one gpu? Are those also being used? How?
What is the inference speed on that ?
I'll do some tests soon.
Could you make a similar video, but with fooocus? 🤩
Aguardando pela disponibilidade da legenda 🙏🏾
Esperamos estar disponível em breve. 👍
@@AIAnimationStudio Você fala português?
@@breno6048 Infelizmente não. Mas a IA e/ou o Google Translate ajudam. Adoraria aprender português um dia.
Hey, Do you know how to install more models on the machine? Thank you for the tutorial
Thank you for this tutorial! Can I just install Comfy + Manager, then paste my workflow and download the missing custom nodes and models trough the manager?
Same question here. Anyone please ?
I found that once I opened up ComfyUI, I was able to 'Load' one of my saved workflows, and use the manager to install any missing custom nodes. If you need help transferring other files like LoRAs or clip_vision files that you can't just get from the manager, I made an extensive reply in this comment chain detailing a process that worked for me for doing this.
Amazing video man.... but do I need to be in the root directory to install and run comfy UI, as I am using sagemaker studio lab and I dont have permission to root directory. Is there no other way to install it without accessing the root directory.
would it be possible to create an image with all dependencies etc. already installed?
Yeah I believe so. At leat one with ComfUI manager would be helpful, so you cna then use that to install any missing dependencies depending on your loaded json workflow.
Is this only intended to work as an administrator user @aianimation? I installed it under the admin account (with SUDO) but when a standard user tries to use it, only the background comes up with none of the nodes
heyy!! just a problem. Everything was working till one week ago, now when I try to open the gui port, it doesn't open the comfyui interface but the jupyter notebook again, some advice?
Hey, somene else hit the same issue. They found a solution. I've pinned their comment (and replies) to the top of the comments. Good luck 👍
@@AIAnimationStudio thankss!! 🤍
Wau very helpful tutorial! So after you download every model and setup everything, is it saved automatically in storage for next instances ?
Unfortunately not. It's saved to that instance. They're working on ways to have possible to move to other instances (not sure how that will work though). So at the moment if you change instances you have to load everything which only takes around 10 mins. (But if you stop and start the same instance (*assuming you can) the files will be there for you.
@@AIAnimationStudio This is what I wanted to know....keeping the file and not deleting like Colab.
@@AIAnimationStudio Loading everything takes 10mins? I'm currently installing the setup and some files are taking over an hour to download! I imagine it's going to take me 5 or 6 hours to get the whole thing set up. Can you explain what you mean by it only taking 10 mins - is it the first time that takes ages but changing instances (I assume that means different GPU configs) only takes 10 mins?
@@Richie_W … no, it shouldn’t take that long by to download the models or anything. From starting from scratch to up and running should be around 10-15 mins . Perhaps the internet speed on the system your renting is crazy slow?? Hmm.
I’ll post a new video soon with some other options … if/when I get time.
And off course it doesnt work for me ;P
Fails at loading...unable to find the image cuda_11.8.0-runtime-ubuntu22.04
Is that after going through all the steps? i.e. can you open Jupyter?
With Comfy setup and running the last bit of code, should load up Comfy UI and make it available via the connecting/open button.
@@AIAnimationStudio fails here... ua-cam.com/video/PvW0VbBK7CY/v-deo.html ... on loading, doesnt install it because it cant find it.
@@AIAnimationStudio Its all working great now. I think the instance I picked just had network problems, another one worked great.
Thanks heaps for the tut. Once I understood whats going on its not to expensive for me (and Im poor) so I can have a play now.
It failed on first run, but I figured it just didnt down a model right and regot it, your script is so useful.
Thanks!
upscale model github link doesnt seem to work for me
Hello i followed the tutorial but now that everything is installed i just cant find the file to actually open comfyUI
whe i hit "open" it opens the folder structure, and there i can´t find the traditional "run_nvidia_:gpu" to open it
in my case that Jupyter only opening after all setup