garystafford

import torch

print(torch.__version__) # e.g., 2.0.0 (at the time of the post)
print(torch.cuda.get_device_name(0)) # e.g., NVIDIA A10G

garystafford

pipeline = DiffusionPipeline.from_pretrained(
    model_name_base,
    torch_dtype=torch.float16,
).to(device)

# new LoRA weights from fine-tuning process
pipeline.load_lora_weights(
    project_name, 
    weight_name="pytorch_lora_weights.safetensors"
)

garystafford

subject_prompt = """oue, photo of a oue electric scooter in a brightly colored 
neon-lite city at night, sleek design, smooth curves, colorful, nighttime, 
urban environment, futuristic cityscape"""

subject_negative_prompt = """person, people, human, rider, floating objects, daytime, 
sunlight, text, words, writing, letters, phrases, trademark, watermark, icon, logo, 
banner, signature, username, monochrome, cropped, cut-off, patterned background"""

refiner_prompt = """ultra-high-definition, photorealistic, 8k uhd, high-quality, 
ultra sharp detail"""

garystafford

subject_prompt = """oue, photo of a oue electric scooter, sleek, smooth curves, colorful, 
daytime, urban, futuristic cityscape"""

subject_negative_prompt = """person, people, human, rider, floating objects, text, 
words, writing, letters, phrases, trademark, watermark, icon, logo, banner, signature, 
username, monochrome, cropped, cut-off, patterned background"""

refiner_prompt = """ultra-high-definition, photorealistic, 8k uhd, high-quality, 
ultra sharp detail"""

garystafford

pipeline = DiffusionPipeline.from_pretrained(
    model_name_base,
    torch_dtype=torch.float16,
).to(device)

pipeline.load_lora_weights(
    project_name, 
    weight_name="pytorch_lora_weights.safetensors"
)

garystafford

!autotrain dreambooth \
    --model ${MODEL_NAME} \
    --project-name ${PROJECT_NAME} \
    --image-path "${IMAGE_PATH}" \
    --prompt "${INSTANCE_PROMPT}" \
    --class-prompt "${CLASS_PROMPT}" \
    --resolution ${RESOLUTION} \
    --batch-size ${BATCH_SIZE} \
    --num-steps ${NUM_STEPS} \
    --gradient-accumulation ${GRADIENT_ACCUMULATION} \

garystafford

import os

# project configuration
project_name = "mb_amg_gt_oue_dreambooth"
model_name_base = "stabilityai/stable-diffusion-xl-base-1.0"
model_name_refiner = "stabilityai/stable-diffusion-xl-refiner-1.0"

# fine-tuning prompts
# 'oue' is a rare tokens, 'car' is a class
instance_prompt = "photo of oue car"

garystafford

%%sh

export PIP_ROOT_USER_ACTION=ignore

pip install -Uq pip # optional
pip install -Uq autotrain-advanced
pip install -q ipywidgets==7.8.1

garystafford

from diffusers import DiffusionPipeline, StableDiffusionXLImg2ImgPipeline

device = "cuda" # cpu or cuda

pipeline = DiffusionPipeline.from_pretrained(
    model_name_base,
    torch_dtype=torch.float16,
).to(device)

refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(

garystafford

# optional descriptive terms: sleek, futuristic, metallic, colorful, urban
subject_prompt = """oue, marker rendering of oue electric scooter, concept art, 
futuristic cityscape, solid color background, bright vibrant colors, marker, sketch, 
illustration, illustrative, marker drawing, expressive strokes, graphic"""

subject_negative_prompt = """person, people, human, rider, floating objects, text, 
words, writing, letters, phrases, trademark, watermark, icon, logo, banner, signature, 
username, monochrome, cropped, cut-off, patterned background"""

refiner_prompt = """sharp, crisp, in-focus, uncropped, high-quality"""