How to DREAMBOOTH your Face in Stable Diffusion. Detailed Tutorial. Best Results.

We use Stable Diffusion Automatic1111 and the DreamBooth extension to fine-tune a custom model using a set of photos. Learn about the installation, the parameters and the best approach as well as possible pitfalls. This solution works locally with AMD and NVIDIA GPUs.

The DreamBooth extension for Automatic1111:
1500 class images for 'person':

Chapters:
0:00 Introduction
2:11 Installing DreamBooth
3:01 Preparing a model
8:13 Instance, class and samples
21:3 The parameters
37:47 Optimizing VRAM
39:47 The training process
43:47 Sample results
45:31 Testing checkpoints
49:30 Text2Image example

#dreambooth #automatic1111 #stablediffusion

[NextTechandAI]

Which (custom) model do you use for fine-tuning?

[WallyHojo]

The Dreambooth options are grayed out. I am unable to select anything.

[AlexFernandoFerreira]

the best tutorial on the topic i have see so far

[Anarky-]

The reason your steps doubled is because you used class images. Every training image is doubled up because it adds a class image to regularize

[eugenekhristo]

Did you try to train LoRA? Is it possible to get such a quality results with trained LoRA as with fine-tning checkpoint model? Thanks

[JojoBenta]

nice tutorial! I am stuck at the begining at3:49... I've installed dreambooth but I can't click on the "create" thumbnail... Maybe it's because I am using the Flux Dev checkpoint. Is there any solution to that? Thanks.

[cedrix]

Hey, thanks for your tuto ! I have a problem when I want to start the training, I have an error because I don't have a in my model file that I created. But instead I have a diffusion_pytorch_model.bin file. I created my model like in your video, do you have also a .bin file in model_name/working/vae ? Or maybe do you know this error ? I also tried to change the .bin by .safetensors but after he is unable to load weights from checkpoint file. Thanks :)

[EngineerBeliefs]

Can we use the same approach for training on objects like a book/jewelry/etc for example?

Or Dreambooth is only for faces? 😊

[ichwill]

Just refuses to work. Think I'll just wait for a better alternative. Too much headache.

[AygyunAhmed-xohg]

I guess the last model of st diffusion includes that dreambooth extension in default.

[EngineerBeliefs]

By the way, can you please make a tutorial on how to train on objects? I tried training on a book (3 images only) and sd generated 300 images of books as class images (they were full of deformities 😢)
And unfortunately the model training is not as good as it was for the face. Even DALLE at that point generated more likewise pictures. So apparently I’m doing smth wrong 🤷‍♀️

[Flash]

Nice to see Dreambooth is still as janky as ever.

[digitalspacestudio]

Thank you, you're a cool guy who taught me.

[martinkaiser]

Welche Auflösung haben dein Trainingphotos ? (die 7 ) Müssen die 512x512 sein ?

[digitalspacestudio]

Hi! I trained the model everything is super! She looks a lot like me. But when I try to change my surroundings or clothes through promt, she doesn't do it. The face remains, but nothing changes around. It doesn't work. If you twist the settings, the environment changes, but I no longer look like it. Am I missing something? Help please

[NCWbu]

What video card are you using in this video?

[strawberryAI_Community]

Exception training model: 'Expected is_sm80 || is_sm90 to be true, but got false. (Could this error message be improved? If so, please report an enhancement request to PyTorch.)'.

[nochan]

What are the parameters if I want to train on a SDXL Checkpoint?

[CodeMania-ye]

In Google collab I just trained my custom model, in 1600 training steps. it's done successfully but when I generate my photo it's looks ridicules. My eyes not perfect or sometimes generate double face on one image. I use google collab because it's take short time, because my gpu is gtx 1050ti 4gb memory, that's why When I use dreembooth on my stable diffusion it's take more than 24 hours show. So, can you tell me how to generate accurate face ?

[CodeMania-ye]

Exception training model: 'Error no file named model_index.json found in directory