metadata
dataset_info:
features:
- name: image
dtype: image
- name: caption
dtype: string
splits:
- name: train
num_bytes: 7421503
num_examples: 10
download_size: 7423067
dataset_size: 7421503
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Nier2b Dataset
Description
The dataset contains images and corresponding captions, which were used to fine-tune the diffusion models.
Usage
To use this dataset for DreamBooth LoRA training, follow the instructions provided in the diffusers library. Here is an example of how to use this dataset:
export MODEL_NAME="black-forest-labs/FLUX.1-dev"
export OUTPUT_DIR=<your output directory>
accelerate launch /diffusers/examples/dreambooth/train_dreambooth_lora_flux.py \
--pretrained_model_name_or_path=$MODEL_NAME \
--output_dir=$OUTPUT_DIR \
--mixed_precision="bf16" \
--instance_prompt="nier2b" \
--dataset_name="Pain-Killer/nier2b-dataset" \
--caption_column="caption" \
--resolution=512 \
--train_batch_size=1 \
--guidance_scale=1 \
--gradient_accumulation_steps=4 \
--optimizer="prodigy" \
--learning_rate=1. \
--report_to="wandb" \
--lr_scheduler="constant" \
--lr_warmup_steps=0 \
--max_train_steps=500 \
--seed="0" \
--validation_prompt="nier2b" \
--validation_epochs=100 \
--push_to_hub
Citation
If you use this dataset in your research, please cite the link to this dataset.