Julia Garner Lora Flux

Prompt

Training With LoRA: The image presents a portrait of Julia Garner with styled, curly hair in an updo. She's wearing a black-and-white sleeveless top, cropped to show only her upper body. The background is a stark white, indicative of a studio or indoor setting, isolating the subject and emphasizing her presence. The composition is clean and uncluttered, with no other elements visible. This deliberate simplicity, combined with the focus on Julia Garner's features, implies a professional or artistic intent, likely intended to highlight her appearance and style.

Prompt

Training Without LoRA: The image presents a portrait of Julia Garner with styled, curly hair in an updo. She's wearing a black-and-white sleeveless top, cropped to show only her upper body. The background is a stark white, indicative of a studio or indoor setting, isolating the subject and emphasizing her presence. The composition is clean and uncluttered, with no other elements visible. This deliberate simplicity, combined with the focus on Julia Garner's features, implies a professional or artistic intent, likely intended to highlight her appearance and style.

Prompt

Testing With LoRA: Julia Garner lying on a bed, almost white bedding; Bali; relax, read a BAZAAR magazine; Photo by Alexandr Tikhomirov; detailed touches; dressed beautiful pretty face with roughly skin, pretty clothes, levity; ultra realistic photo --q 2 --s 750 --v 5 --chaos 28 --v 5 --q 2

Prompt

Testing Without LoRA: Julia Garner lying on a bed, almost white bedding; Bali; relax, read a BAZAAR magazine; Photo by Alexandr Tikhomirov; detailed touches; dressed beautiful pretty face with roughly skin, pretty clothes, levity; ultra realistic photo --q 2 --s 750 --v 5 --chaos 28 --v 5 --q 2

All files are also archived in https://github.com/je-suis-tm/huggingface-archive in case this gets censored.

This a non-quantized version of https://huggingface.co/je-suis-tm/julia_garner_lora_flux_nf4. Both are trained on the same dataset. The training is based on https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/test_dreambooth_lora_flux.py. The training took 3 hours on A100 80GB with max VRAM consumption at 35GB. The inference consumes 36GB of VRAM.

Train

export MODEL_NAME="black-forest-labs/FLUX.1-dev"
export INSTANCE_DIR="/pvol/Julia Garner"
export OUTPUT_DIR="/pvol/julia_garner_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="Julia Garner" \
  --caption_column="text" \
  --resolution=1024 \
  --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

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/julia_garner_lora_flux",
                       weight_name='pytorch_lora_weights.safetensors')


prompt = "Julia Garner lying on a bed, almost white bedding; Bali; relax, read a BAZAAR magazine; Photo by Alexandr Tikhomirov; detailed touches; dressed beautiful pretty face with roughly skin, pretty clothes, levity; ultra realistic photo --q 2 --s 750 --v 5 --chaos 28 --v 5 --q 2"

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("Julia Garner.png")