[Stable Diffusion] Stable Diffusion 3.5 Text to Image 이미지 생성

[Stable Diffusion] Stable Diffusion 3.5 Text to Image 이미지 생성

테스트 환경

Mac m1 pro

Code

import os
import uuid
from datetime import datetime

import torch
from diffusers import StableDiffusion3Pipeline
from dotenv import load_dotenv

load_dotenv()
HUGGING_FACE_ACCESS_TOKEN = os.getenv("HUGGING_FACE_ACCESS_TOKEN")

"""
https://prompthero.com/stable-diffusion-cartoon-prompts
"""

# torch.backends.mps.enable_fallback_implementations = True
print(f"Starting inference... {datetime.now()}")

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    torch_dtype=torch.bfloat16,
    token=HUGGING_FACE_ACCESS_TOKEN,
    local_files_only=True,
    add_prefix_space=True,
    safety_checker=None,
)
# pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe = pipe.to("mps")  # cuda
# pipe.enable_attention_slicing()

# prompt = "A hamster holding a sign that reads Hello World"
# prompt = "happy retriever with smile in the snow"
prompt = """"Cat's Table", A cartoon in the style of John Tenniel, Punch magazine, 1890"""

image = pipe(
    prompt,
    width=512,
    height=512,
    num_inference_steps=28,
    guidance_scale=3.5,
).images[0]
image.save(image_filename := os.path.join("images", f"{uuid.uuid4()}.png"))
print("Image saved to", image_filename, f"Finished inference... {datetime.now()}")

Output