How to train Stable Diffusion on your photos on a remote GPU: using RunPod and Dreambooth

Chapters:
0:00 Intro
7:00 Sign up for RunPod
14:42 Clone Stable Diffusion
22:14 Download Stable Diffusion
27:01 Training Images
31:01 Start Training
35:29 Troubleshooting
39:29 Project name
40:09 Generating Images
44:49 Conclusion

To download Stable Diffusion model:
https://huggingface.co/

RunPod:
https://runpod.io?ref=bec9byrf
(if you download following this link, I may get some free rented hours on RunPod, you can get a referral link on the website too)

Helper:
https://www.kris.art/runpoddreambooth

Resize:
https://www.birme.net/

Convert JPG to PNG:
https://www.iloveimg.com/jpg-to-image

Celebrity alike:
https://www.starbyface.com/

Find great prompts:
https://lexica.art/

33 Comments

  1. Great tutorial, there is a way to download the checkpoint file straight to your google drive, and its just as fast as uploading it to runpod. I need to watch how to doit again.

  2. Spent 2 nights to do this tutorial (first an electrical failure ruined it) and today I have this message when I launch the training (32 minutes of the video) : "NameError: name 'dataset' is not defined"
    impossible to go further… so frustrating and time consuming. If anyone has a clue..

  3. Your videos are miraculous! They managed to get me into the mysterious mechanisms of runpod with ease … Thanks!
    If I may I would like to ask you a question: I have seen that on Runpod there is no latest version of SD AUTOMATIC 1111 installed. Do you have to suggest a solution to upgrade? And then to install the sd 1.5 version of a few days ago? I thank you.

  4. Thanks for this tutorial, Kris ! I have been following your adventures on the SD/formerly MJ facebook group for a while now, and I've learned many things from you 🙂

  5. Great tutorials! Always have success when following your steps. Could you possibly do a video on combining multiple trained models? As in one that will incorporate two separate tokens, for example one used to create my portait as well as another specified person or object? Thanks! I'll keep watching and learning!

  6. Helllo and thank's for the tuto
    i got an issue while i start training, during the processs i Game of get : ValueError("num_samples should be a positive integer "
    ValueError: num_samples should be a positive integer value, but got num_samples=0
    or expected for clipboard or unexepcted
    thx

  7. Hi! this video is getting a bit confusing at the google drive part. First, in the video, the gdown command u used has an id, on your website helper, there is no id. I tried to copy paste the no-id version of the command, it is not downloading anything. You mentioned there is some model in the google drive, is there a previous video that should be watched before watching this?

  8. Thank you so much for doing this video. I've been making models for awhile with bad results. I was using the celebrity suggestion but I never realized I was putting the celebrity name as the class instead of the token. I was doing this wrong until I saw you do it the correct way. Now I have a great model. Thank you, you are an angel!

  9. Hello! Nice tuto! I use you tips to train my model using my wife photos. There is any way i run the new Dreambooth with Prior Preservation? It comes integrated with the SD Web UI on Runpod but it fails when i tried to run Train, but with no error logs, do you just run this new version on Runpod?

  10. Is this still the best way to do this given all the updates? For example dreambooth now has optimization and such, so much so that I can run it on my CPU locally! (It will take 4 and a half days though lol)

    Thanks for any advice!

  11. Thank you so much for this. You are great a great teacher!! I have successfully trained myself and my wife. I would love to train multiple models and not have to start from the beginning. I try and change out my training images and token name and file name but I keep getting errors when I click the start training section. Have you tried to train different people in the same runpod without having to start from the beginning? Seems like I could save a lot of time if I didn't need to upload the model, download the regulization images etc.. each time. Thank yhou so much for all your help. You rock!

Leave a Reply

Your email address will not be published. Required fields are marked *

© 2024 AI Art Video Tutorials