Epic Web UI DreamBooth Update - New Best Settings - 10 Stable Diffusion Training Compared on RunPods

Playlist of #StableDiffusion Tutorials, #Automatic1111 and Google Colab Guides, #DreamBooth, Textual Inversion / Embedding, LoRA, AI Upscaling, Pix2Pix, Img2Img:

The experiments are done on RunPods but if you have 20 GB vram having Windows PC, you can exactly do the same things. All you need to do is installing latest Automatic1111 and DreamBooth extension.

Where you can find executed commands and prompts and more info:

2400 Photo Of Man classification images:

DreamBooth extension repo link:

Easiest Way to Install & Run Stable Diffusion Web UI on PC by Using Open Source Automatic Installer:

How to use Stable Diffusion V2.1 and Different Models in the Web UI - SD 1.5 vs 2.1 vs Anything V3:

Zero To Hero Stable Diffusion DreamBooth Tutorial By Using Automatic1111 Web UI - Ultra Detailed:

How To Do Stable Diffusion Textual Inversion (TI) / Text Embeddings By Automatic1111 Web UI Tutorial:

Sketches into Epic Art with 1 Click: A Guide to Stable Diffusion ControlNet in Automatic1111 Web UI:

Ultimate RunPod Tutorial For Stable Diffusion - Automatic1111 - Data Transfers, Extensions, CivitAI:

Transform Your Selfie into a Stunning AI Avatar with Stable Diffusion - Better than Lensa for Free:

Stable Diffusion Google Colab, Continue, Directory, Transfer, Clone, Custom Models, CKPT SafeTensors:

Fantastic New ControlNet OpenPose Editor Extension & Image Mixing - Stable Diffusion Web UI Tutorial:

0:00 Introduction to DreamBooth new update best settings experiments
0:49 How to setup a new Pod and install DreamBooth newest update properly
2:25 New RunPod started time to setup and install DreamBooth
3:20 Install DreamBooth extension manually and fix errors
7:15 Starting first experiment test0 - setup of settings
7:35 Best DreamBooth settings for 12GB VRAM having GPUs
11:05 Setting and starting second experiment test1 DEIS Noise Scheduler
11:55 Test2 Unfreeze Model
12:12 Test3 Lion Optimizer
12:40 Test4 Stable Diffusion Offset Noise
13:17 Test5 Freeze Clip Normalization Layers DreamBooth
13:33 Test6 Use EMA + Use EMA for Prediction
14:37 Test7 Use EMA + Use EMA Weights for Inference
14:55 Test8 Use EMA only
15:05 Solution of configuration index out of range Web UI error
15:18 Test9 Don't use xformers - default memory attention and fp16
15:49 How to add more disk space to your existing RunPod
17:08 xformers related bug error
18:20 How to continue DreamBooth training if an error occurs or for any reason halted
18:49 All tests have been completed time to check their training samples
18:58 Test0 training samples - previously known best settings for 12GB VRAM
19:25 Test1 training samples - DEIS Noise Scheduler
19:44 Test2 training samples
20:20 Test3 training samples
21:13 Test4 training samples
21:38 Test5 training samples
22:15 Test6 training samples
23:27 Test7 training samples
24:19 Test8 training samples
24:54 Test9 training samples
25:36 Finding a good seed to compare all checkpoints within each trained model
26:46 What seed and prompt I used to compare checkpoints
27:50 What is the logic of starting seed when using batch image generation
28:14 How to use x/y/z plot to compare checkpoints to find best trained model
29:10 Where to find x/y/z plot generated grid image file
29:40 Comparing generated grid files of all experiments
29:44 Test0 Checkpoints Grid
31:26 Test1 Checkpoints Grid DEIS Noise Scheduler
32:50 Test2 Checkpoints Grid
34:48 Test3 Checkpoints Grid
38:19 Test4 Checkpoints Grid
40:30 Test5 Checkpoints Grid
41:25 Test6 Checkpoints Grid
43:14 Test7 Checkpoints Grid
44:00 Explanation of overtraining by a comparison
45:27 Test8 Checkpoints Grid
46:59 Test9 Checkpoints Grid
48:30 How to download all decided best checkpoints via runpodctl
49:09 What is RunPod connect to web terminal and how to use it when jupyter connection is not available
49:56 Where to put downloaded safetensors model files
50:10 Using x/y/z plot do conduct final comparison experiment on my local web UI
50:40 Test7 model file size is smaller than others
51:00 Each model file sizes
51:05 Comparing all of the experiments test0 vs test1 vs test2 ...
1:00:20 Ending Speech

[SECourses]

Please join discord, mention me and ask me any questions. Thank you for like, subscribe, share and Patreon support. I am open to private consulting with Patreon subscription.

[AliShahmeer]

It feels illegal to watch this for free, You’re such a legend for spending so much time going over each and every detail and going through the pain to caption the video. Huge respect for you sir!

[Sergiosvm]

At first I wasn't a fan because of your english pronunciation, making it hard to understand you sometimes, but you truly make great stable diffusion content and I really appreciate it!

[___x__x_r___xa__x_____f______]

super video. Real methodolical approach. We need more of this. Thanks

[freke]

Wow, this video was very informative! I learned a lot about the new features of Web UI DreamBooth and how to use them. You did a great job of explaining and demonstrating everything in a clear and engaging way. Thank you for sharing your knowledge and experience with us. I’m looking forward to more videos from you! :)

[HughOBrien]

Incredible resource! You saved me a lot of time experimenting!

[AndyGilleand]

Do you have any recommendations for training a 2.1 x768 model on a 12GB card? Is it possible with any settings?

[ookpalm]

Your videos helped me a lot. Thank you for making tutorial videos for us! Couldn't get it up and running without your resources

[pastuh]

If I wanted to perform fine-tuning for any stable-diffusion model, how many times could I input a similar object?
In other words, would the model retain all the provided images, or would it overwrite them after some time?
Question about model, not about Lora. Thanks

[bendito]

Awesome video you put in the incredible work!
So in summary it sounds kind of like
Lion - probably good but make sure to adjust learn rate divide by 10
Use ema (but not ema prediction or inference) -good, as usual, but good to know not to be tempted to turn on those other 2 ema settings.
Not using Xformers, good thing if enough vram. I have 24gb vram Tesla m40 so I don't use Xformers.

Takeaway is the Lion optimizer is interesting and I'll definitely try it out myself.

[iforme]

thanks for the amazing video
could you please make a video of training a style of painting
i m really strugling with training styles

[nubye]

I noticed most of your training videos uses checkpoints rather than Loras. Do you suggest Loras too as the file sizes are smaller and do you have a good video on its settings?

[SilentD]

what is it that triggers your ability to check for "stylization" is that your prompt in the sanity sample box "by thomar hanuka"? I also cant get any trained sample images to look any decent with 1.5. They all look distorted bad faces. what am I doing wrong to always have bad renders with 1.5?

even creating class photos of : photo of a woman all the pictures look horribly bad.

Using kohya creates much better results than the built in dreambooth extention. but there are no previews in kohya so I feel like im running blind. I may have something wrong in my system? do I need a yaml file for 1.5 like the 2.1 uses?

[kaidenquinart]

how many step for teaching style? my dataset is 100 image and i have 1000 class image

[MavVRX]

Does increasing the number of subject training images (without increasing training epochs) give a better or worse result?

[radry]

I tried to follow this guide and set up dreambooth but when creating a new model it will crash/terminate when "extracting unet"

[TransformXRED]

For your test 3, you changed the training rate to a lower one when you selected the Lion optimiser.

[ss-brtq]

Wonderful content, which let me uncover a lot of confusion, thank you!
As for training pictures, I found that you didn't add a caption txt file. Did you skip this step in the video or can you get good training results without adding that file? To be honest, it's too troublesome to add and modify the picture description😅

[donutello_]

i hope thelastben gets wind of this, he does the google colab version of dreambooth for us peasants haha

[tariksaid]

hello thank you fot the amazing video
could please make a video about kohya ss on Runpod
i just can t make it work on Runpod