ComfyUI - RunPod: Installation and use tutorial Cheap rental GPU ComfyUI Manager Stable Diffusion

Поділитися
Вставка
  • Опубліковано 2 гру 2024

КОМЕНТАРІ • 44

  • @astralboy42
    @astralboy42 29 днів тому +2

    Thanks so much for this! I've tried tackling Runpod before and it was a nightmare but with your image its been very slick and easy! Thanks so much!

  • @AshT8524
    @AshT8524 14 днів тому

    That was very useful, I'm planning to rent a GPU from runpod and this video answered all my questions. Thank you

    • @koalanation
      @koalanation  12 днів тому

      Glad to hear is being useful! When you start knowing the machines which are available in each location, learn how to plan and fit what you rent and what your goal is.

  • @az3848
    @az3848 11 днів тому

    Thanks for the great tutorial! Is it possible to install forge or swarm on the same network volume to save space?

    • @koalanation
      @koalanation  11 днів тому +1

      Yes, it is possible, but probably you need to change some configuration file (extra_model_paths.yaml). Check on the comfyui documentation

  • @RaggazzasMarketing
    @RaggazzasMarketing Місяць тому +1

    Hi, I've used the template but it doesn't have the ComfyUI installed... the template is empty

    • @koalanation
      @koalanation  Місяць тому +1

      Hi! The template uses a pytorch container, and then runs a script which downloads and installs ComfyUI and ComfyUI Manager. This takes some time, but after few minutes you should be able to run run_gpu.sh in /workspace

    • @RaggazzasMarketing
      @RaggazzasMarketing Місяць тому

      Thanks for answering, I had to run the Install sh file first to install everything again because the run_cpu and run_gpu files were missing, then after everything was re-installed and the run files created, I ran the run_gpu.sh file and it loads ok.. but I can't access the web with ComfyUI because it never activates in the connections in the pod option

    • @koalanation
      @koalanation  Місяць тому

      in the pod, make sure ports 8888 and 8188 are open. If they are not, click on 'More actions' (the 3 lines icon) and click on 'edit pod'. Then you can change the open ports there (if will ask to restart the pod).
      Make sure also that in run_gpu.sh you have the last line 'python main.py --preview-method auto --listen'.

  • @TimelessMythology
    @TimelessMythology Місяць тому

    The tutorial video is very detailed; I'm truly grateful for that. Can I upload files from my computer to Jupyter?

    • @koalanation
      @koalanation  Місяць тому

      Yes. In Jupyterlab you should have the possibility to navigate through the folders. Uploading you do by simply drag your local file and dropping it on the folder/directory. You can download the files from the networks storage by right clicking and downloading the file or directory you want.

  • @anjanagajanayake7685
    @anjanagajanayake7685 4 дні тому

    can this method work for a flux gguf workflow as well?

    • @koalanation
      @koalanation  4 дні тому

      You can use it for flux or any other workflow for comfyui.

  • @heywarneywarney6959
    @heywarneywarney6959 2 дні тому

    Btw why is the checkpoint folder restricted and only accessible from the terminal??

    • @koalanation
      @koalanation  День тому

      @@heywarneywarney6959 I believe it has to do Jupiter restricts checkpoints as a word for saved notebooks or something like that.

  • @DanielPartzsch
    @DanielPartzsch Місяць тому

    Thank you. Is it possible that multiple different users can share the same installation of comfyui or at least the network volume while still be able to work in parallel on individual render nodes? So basically my team mates can all access the same models stored in one location instead of uploading them indivually per user?

    • @koalanation
      @koalanation  Місяць тому +1

      You can convert your account to team account. There is an option on the left menu in runpod

  • @heywarneywarney6959
    @heywarneywarney6959 2 дні тому

    How to delete a file in the checkpoints folder if it's not accessible normally?

    • @koalanation
      @koalanation  День тому

      @@heywarneywarney6959 Unfortunately, with jupyter you have to do it from the command line. In the workspace folder, open a terminal and go to the checkpoints directory typing "cd ComfyUI/models/checkpoints". Then, type "rm -r filename.ckpt" (or the name of the file you want to delete)

  • @PoolsideTobago
    @PoolsideTobago День тому

    Hi i'm getting an ai dock login screen for each connect after script has installed on runpod, any advice please.

    • @koalanation
      @koalanation  День тому

      @@PoolsideTobago are you using this template? runpod.io/console/deploy?template=mbdoaf7ea1&ref=617ypn0k

    • @koalanation
      @koalanation  День тому

      @@PoolsideTobago or this for the pytorch 2.2 version runpod.io/console/deploy?template=3c0tdgrly3&ref=617ypn0k

  • @mu5237
    @mu5237 Місяць тому

    It was soooo helpful. Thanks a lot!
    Just one question, how do I delete the ckpt file in a checkpoint folder if I can't view the folder?

    • @mu5237
      @mu5237 Місяць тому

      I've downloaded same cptk file several times. So I would like to leave just one and delete the rest

    • @koalanation
      @koalanation  Місяць тому +2

      Unfortunately, with jupyter you have to do it from the command line. In the workspace folder, open a terminal and go to the checkpoints directory typing "cd ComfyUI/models/checkpoints". Then, type "rm -r filename.ckpt" (or the name of the file you want to delete)

    • @mu5237
      @mu5237 5 днів тому

      @@koalanation Thank you so much!

  • @DanishShabbir-u6k
    @DanishShabbir-u6k 20 днів тому

    Great Video!
    I have a question here what will be best approach to host workflows as API (I know comfy UI has API interface)!
    Like I want to publically host these so it can receive request from my front end app for Public!

  • @Blueclaws85
    @Blueclaws85 Місяць тому

    After getting to the section where you switch over to ComfyUI, it seems RunPod doesn't detect a service is running? Any suggestions?

    • @koalanation
      @koalanation  Місяць тому

      Check on the log if the installation of comfyuI is complete. A message about the pod being ready to use should appear.

  • @TUSHARGOPALKA-nj7jx
    @TUSHARGOPALKA-nj7jx Місяць тому

    How to save the workflow on the network volume disk

    • @koalanation
      @koalanation  Місяць тому +1

      The saved workflows are in "workspace/ComfyUI/user/default/workflows"

  • @cheehaowong1714
    @cheehaowong1714 2 місяці тому

    hi, why I couldn't find the template in runpod?

    • @koalanation
      @koalanation  2 місяці тому +1

      Can you please check: www.runpod.io/console/explore/3c0tdgrly3

    • @cheehaowong1714
      @cheehaowong1714 2 місяці тому

      @@koalanation thanks!

  • @alecubudulecu
    @alecubudulecu Місяць тому

    Jupyter lab requires password and token. They have info on it but was hoping you’d cover that.

    • @koalanation
      @koalanation  Місяць тому

      Normally the password is sent as a environment variable and it is not needed to start jupyterlab. In case is needed, when the pod has started can be found seen in 'edit pod' under environment variables 'JUPYTER_PASSWORD'

  • @faycaltech5922
    @faycaltech5922 Місяць тому

    i dont see jupyter lab when i connect

    • @koalanation
      @koalanation  Місяць тому

      @@faycaltech5922 should appear in port 8888 after everything is installed

  • @Stef_frenchtouch
    @Stef_frenchtouch Місяць тому

    any other solution .. because run-of there are cost when start and stop .. if it's stop tool long there are cost and must be start all day with 0,0023 cost by hour .. I search an option for only paid when I do a renderer . I have a Mac M1 with flux dev difficult with 40min for one pic .. 😂

    • @koalanation
      @koalanation  Місяць тому

      @@Stef_frenchtouch there are only render solutions but they charge more by the render time, as they need to keep the machines hot. Look for what modal can do for you, for example, or if you can do runpod serverless.

    • @Stef_frenchtouch
      @Stef_frenchtouch Місяць тому

      @@koalanation yes but an instance who’s cost by hour is more than cost by only renderer ! Replicate is perfect whit api for example but to limited fir comfyUI use ( not Lora , not upload personal Lora etc ..)

    • @koalanation
      @koalanation  Місяць тому

      @@Stef_frenchtouch I am not very familiar with replicate, but I think if you want to personalize more your execution (use different models than the standard ones) you need to set your own deployment.