Multi Subject DreamBooth training

DreamBooth is a method to personalize text2image models like stable diffusion given just a few(3~5) images of a subject. This train_multi_subject_dreambooth.py script shows how to implement the training procedure for one or more subjects and adapt it for stable diffusion. Note that this code is based off of the examples/dreambooth/train_dreambooth.py script as of 01/06/2022.

This script was added by @kopsahlong, and is not actively maintained. However, if you come across anything that could use fixing, feel free to open an issue and tag @kopsahlong.

Running locally with PyTorch

Installing the dependencies

Before running the script, make sure to install the library's training dependencies:

To start, execute the following steps in a new virtual environment:

git clone https://github.com/huggingface/diffusers
cd diffusers
pip install -e .

Then cd into the folder diffusers/examples/research_projects/multi_subject_dreambooth and run the following:

pip install -r requirements.txt

And initialize an 🤗Accelerate environment with:

accelerate config

Or for a default accelerate configuration without answering questions about your environment

accelerate config default

Or if your environment doesn't support an interactive shell e.g. a notebook

from accelerate.utils import write_basic_config
write_basic_config()

Multi Subject Training Example

In order to have your model learn multiple concepts at once, we simply add in the additional data directories and prompts to our instance_data_dir and instance_prompt (as well as class_data_dir and class_prompt if --with_prior_preservation is specified) as one comma separated string.

See an example with 2 subjects below, which learns a model for one dog subject and one human subject:

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export OUTPUT_DIR="path-to-save-model"

# Subject 1
export INSTANCE_DIR_1="path-to-instance-images-concept-1"
export INSTANCE_PROMPT_1="a photo of a sks dog"
export CLASS_DIR_1="path-to-class-images-dog"
export CLASS_PROMPT_1="a photo of a dog"

# Subject 2
export INSTANCE_DIR_2="path-to-instance-images-concept-2"
export INSTANCE_PROMPT_2="a photo of a t@y person"
export CLASS_DIR_2="path-to-class-images-person"
export CLASS_PROMPT_2="a photo of a person"

accelerate launch train_multi_subject_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir="$INSTANCE_DIR_1,$INSTANCE_DIR_2" \
  --output_dir=$OUTPUT_DIR \
  --train_text_encoder \
  --instance_prompt="$INSTANCE_PROMPT_1,$INSTANCE_PROMPT_2" \
  --with_prior_preservation \
  --prior_loss_weight=1.0 \
  --class_data_dir="$CLASS_DIR_1,$CLASS_DIR_2" \
  --class_prompt="$CLASS_PROMPT_1,$CLASS_PROMPT_2"\
  --num_class_images=50 \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=1e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=1500

This example shows training for 2 subjects, but please note that the model can be trained on any number of new concepts. This can be done by continuing to add in the corresponding directories and prompts to the corresponding comma separated string.

Note also that in this script, sks and t@y were used as tokens to learn the new subjects (this thread inspired the use of t@y as our second identifier). However, there may be better rare tokens to experiment with, and results also seemed to be good when more intuitive words are used.

Inference

Once you have trained a model using above command, the inference can be done simply using the StableDiffusionPipeline. Make sure to include the identifier(e.g. sks in above example) in your prompt.

from diffusers import StableDiffusionPipeline
import torch

model_id = "path-to-your-trained-model"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")

prompt = "A photo of a t@y person petting an sks dog"
image = pipe(prompt, num_inference_steps=200, guidance_scale=7.5).images[0]

image.save("person-petting-dog.png")

Inference from a training checkpoint

You can also perform inference from one of the checkpoints saved during the training process, if you used the --checkpointing_steps argument. Please, refer to the documentation to see how to do it.

Additional Dreambooth documentation

Because the train_multi_subject_dreambooth.py script here was forked from an original version of train_dreambooth.py in the examples/dreambooth folder, I've included the original applicable training documentation for single subject examples below.

This should explain how to play with training variables such as prior preservation, fine tuning the text encoder, etc. which is still applicable to our multi subject training code. Note also that the examples below, which are single subject examples, also work with train_multi_subject_dreambooth.py, as this script supports 1 (or more) subjects.

Single subject dog toy example

Let's get our dataset. Download images from here and save them in a directory. This will be our training data.

And launch the training using

Note: Change the resolution to 768 if you are using the stable-diffusion-2 768x768 model.

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --instance_prompt="a photo of sks dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=400

Training with prior-preservation loss

Prior-preservation is used to avoid overfitting and language-drift. Refer to the paper to learn more about it. For prior-preservation we first generate images using the model with a class prompt and then use those during training along with our data. According to the paper, it's recommended to generate num_epochs * num_samples images for prior-preservation. 200-300 works well for most cases. The num_class_images flag sets the number of images to generate with the class prompt. You can place existing images in class_data_dir, and the training script will generate any additional images so that num_class_images are present in class_data_dir during training time.

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800

Training on a 16GB GPU:

With the help of gradient checkpointing and the 8-bit optimizer from bitsandbytes it's possible to run train dreambooth on a 16GB GPU.

To install bitandbytes please refer to this readme.

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=2 --gradient_checkpointing \
  --use_8bit_adam \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800

Training on a 8 GB GPU:

By using DeepSpeed it's possible to offload some tensors from VRAM to either CPU or NVME allowing to train with less VRAM.

DeepSpeed needs to be enabled with accelerate config. During configuration answer yes to "Do you want to use DeepSpeed?". With DeepSpeed stage 2, fp16 mixed precision and offloading both parameters and optimizer state to cpu it's possible to train on under 8 GB VRAM with a drawback of requiring significantly more RAM (about 25 GB). See documentation for more DeepSpeed configuration options.

Changing the default Adam optimizer to DeepSpeed's special version of Adam deepspeed.ops.adam.DeepSpeedCPUAdam gives a substantial speedup but enabling it requires CUDA toolchain with the same version as pytorch. 8-bit optimizer does not seem to be compatible with DeepSpeed at the moment.

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch --mixed_precision="fp16" train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --sample_batch_size=1 \
  --gradient_accumulation_steps=1 --gradient_checkpointing \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800

Fine-tune text encoder with the UNet.

The script also allows to fine-tune the text_encoder along with the unet. It's been observed experimentally that fine-tuning text_encoder gives much better results especially on faces. Pass the --train_text_encoder argument to the script to enable training text_encoder.

Note: Training text encoder requires more memory, with this option the training won't fit on 16GB GPU. It needs at least 24GB VRAM.

export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --train_text_encoder \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --use_8bit_adam \
  --gradient_checkpointing \
  --learning_rate=2e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800

Using DreamBooth for other pipelines than Stable Diffusion

Altdiffusion also support dreambooth now, the runing comman is basically the same as abouve, all you need to do is replace the MODEL_NAME like this: One can now simply change the pretrained_model_name_or_path to another architecture such as AltDiffusion.

export MODEL_NAME="CompVis/stable-diffusion-v1-4" --> export MODEL_NAME="BAAI/AltDiffusion-m9"
or
export MODEL_NAME="CompVis/stable-diffusion-v1-4" --> export MODEL_NAME="BAAI/AltDiffusion"

Training with xformers:

You can enable memory efficient attention by installing xFormers and padding the --enable_xformers_memory_efficient_attention argument to the script. This is not available with the Flax/JAX implementation.

You can also use Dreambooth to train the specialized in-painting model. See the script in the research folder for details.