export TMPDIR=/pvol/tmp
        export TRANSFORMERS_CACHE=/pvol/huggingface-cache
        export HF_HOME=/pvol/huggingface-home
        export MODEL_NAME="black-forest-labs/FLUX.1-dev"
        export INSTANCE_DIR="/pvol/marvel_zombies_style"
        export OUTPUT_DIR="/pvol/marvel_zombies_style_lora_flux"
        accelerate config default
        accelerate launch train_dreambooth_lora_flux1.py \
          --pretrained_model_name_or_path=$MODEL_NAME  \
          --mixed_precision="bf16" \
          --dataset_name=$INSTANCE_DIR \
          --output_dir=$OUTPUT_DIR \
          --gradient_checkpointing \
          --instance_prompt="Marvel Zombies Style" \
          --caption_column="text" \
          --resolution=720 \
          --train_batch_size=1 \
          --guidance_scale=1 \
          --use_8bit_adam \
          --checkpointing_steps=100 \
          --gradient_accumulation_steps=4 \
          --optimizer="adamW" \
          --learning_rate=1e-4 \
          --lr_scheduler="constant" \
          --lr_warmup_steps=100 \
          --max_train_steps=1500 \
          --rank=4 \
          --seed="0"

Usage

python

import torch
from diffusers import FluxPipeline

device = "cuda:0"

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.bfloat16)

pipe.load_lora_weights("je-suis-tm/marvel_zombies_style_lora_flux",
                       weight_name='pytorch_lora_weights.safetensors')


prompt = "Marvel Zombies Style, Gothic Fire-Breathing winged Dragon in a stone courtyard being Roaring into the night sky"

image = pipe(
  prompt=prompt,
  generator=torch.Generator(device=device).manual_seed(42),
  num_inference_steps=50, # 28 is a good trade-off
  guidance_scale=4,
  height=1024, 
  width=1024,
).images[0]

image.save("marvel_zombies_style.png")

Trigger words

You should use Marvel Zombies Style to trigger the image generation.

Download model

Download them in the Files & versions tab.