Dreambooth Tutorial: Train Stable Diffusion Image AI With Your Own Model For Less Than a Dollar!

I made a quick tutorial on how to train Dreambooth and Stable Diffusion with your own images on a cloud GPU provider. In this video I go through the steps to prepare for training and go step-by-step through the process, and show a few common issues that you can run into when attempting to train your model. The final cost for the entire training ended up being only $0.35, so this is a very cost effective way to train your own models if you don't have a higher powered GPU and computer.

Resources:
Path for saving to cloud storage: /workspace/Dreambooth-Stable-Diffusion/trained_models

00:00 Intro and Preparation
6:11 Training on Runpod
25:21 Downloading the finished model
32:00 Live demonstration of trained corgi model
28:41 Conclusion and final thoughts

[ideamonkey]

Thank you so much for the great video. : ) The troubleshooting of killing other processes was super helpful too.
Also ran into a small issue, posting here in case it happens to others - got an "image file is truncated" error during the training process. From googling, it seemed likely one of the training images I uploaded was freaking it out. I just deleted images two by two a few times until it worked.

[JoeTheDino]

Great video and beautiful dog ! A lot of people are making blind tutorials on training humans. I understood a lot more watching you training on your doggo. You're making great content, subbed. Please give a big hug to your dog for me.

[monsterwerksvideo]

Thank you for your tutorial. Your slightly deeper dive into runpod was super helpful.

[YokosoOtaku]

This video is absolutely fantastic. Thank you so much for helping me train. It's been a lot of fun and I've had great results. The comment about how to train other models was super helpful too and allowed me to do even more. Really and truly, thank you!

[chadcrypto]

thank you so much for showing how to use runpod with the training.

[audiogus]

Cool cool! thanks for showing how Runpod works, I only ever tried the Runpod web UI and have been baffled about how to approach notebooks on it. Looks a fair bit more accessible now, thanks!

[shaneofgames]

Hey, man you are really funny and have great content! I’ve been programming poorly for a few years and lately started working with AI, so I’m really enjoying your videos. Thanks for the tutorial. Subbed, best of luck.

[switterbeet]

Thanks a lot for this amazing tutorial! This is probably the best tutorial for this out there It worked after some errors tho! Im amazed

[macguy]

AWESOME Video! Exactly what I was looking for. A few questions:
1. Do you have a script for 2.1?
2. What cloud service did you use that allows you to use the model that you created and how much did that cost?

[TheWaTchoTcho]

thank you so much, do you have any idea on how to run multiple token ? i'm following Digital Corridor tutorial

[Andimax]

I followed this tutorial for training my first model and it ended up successful, but now trying to train my second model, when I attempt to run the "training" block, I press run and nothing pops up. It has a "*" next to the training block like its running but no text or output shows up below. Does anyone what this means or how to fix it?

[screenrec]

Thanks for posting. The video is very detailed. With the up coming Gpu like the RTX4090, do you think training will be faster ? I am looking forward to local training. Internet is not that stable for the area where I am living.

[benjavalanche]

This was really helpful. Still had a bit of trouble with the kernel seizing up, but otherwise this was great. Thanks!

[MrRedstoneready]

I had the most trouble pulling the model onto anything like Google Cloud or dropbox, so I just downloaded it locally. maybe 10 minutes and a few cents with good internet

[esmd]

Encountering this error now when I run the code: RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu)

Anyone know what this is?

[OmriSadeh]

This video was super helpful in all perspectives of this process, thank you!
Now that the new 1.5 SD models are out, one of them (the 7gb one) is supposed to yield better results specifically in training dreambooth. do you have any idea on how to switch the model used in this process (1.4) with the new one which is also found on huggingface?
if you could do it on your repo it would be very appreciated

[sahiliqbal]

I'm unable to get cloud sync working with b2. I have a bucket and the correct api keys etc... But when I finalize it, instead of giving me the progress bar on runpod, it just restarts and says please pick an integration (beginning cloud sync page)... Any ideas?

[bladechild]

Whatever you do don't use the community cloud pods. They download and upload at really awful speeds. I trained a model for an hour just to realize that it was gonna take 3 hours for them to upload it to google drive

[ingoliebig]

@Hung Tuong
How many pictures would you recommend to use? Should they be cropped beforehand (512x512?) and should they be mostly the same or different perspective? With background or background removed? Always same lightning or different? Thanks in advance. I am new to this and have been trying to train a modle the past couple days and can't get it to work even when paying Colab

[MrPlasmo]

Great tutorial thanks. You mentioned that there is a way to do this locally for those who have a good gpu without paying for service. I’ve run several training instances using free google colab which sometimes works, but sometimes after hours it quits or kicks you out so am interested in trying to make some training models locally as I want to experiment with various settings and images.

Can you post (or point me to another video) tutorial on how to run training locally? Most videos online have instructions for training either using google colab or a paid gpu offsite. Thanks!