data/diffusion/generate_lustify.py

#!/usr/bin/env python3
"""Generate images with the locally downloaded Lustify SDXL model."""

from __future__ import annotations

import argparse
import os
from pathlib import Path

os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")

import torch
from diffusers import DPMSolverMultistepScheduler, StableDiffusionXLPipeline


DEFAULT_MODEL = Path(__file__).resolve().parent / "models" / "Lustify-SDXL-GGWP-v7"
DEFAULT_NEGATIVE_PROMPT = (
    "low quality, worst quality, blurry, deformed, distorted, bad anatomy, "
    "bad hands, extra fingers, missing fingers, watermark, text, logo"
)


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Generate an image with Lustify SDXL GGWP v7")
    prompt_group = parser.add_mutually_exclusive_group(required=True)
    prompt_group.add_argument("--prompt", help="Text prompt")
    prompt_group.add_argument(
        "--prompt-file",
        type=Path,
        help="Read the text prompt from a UTF-8 file",
    )
    negative_group = parser.add_mutually_exclusive_group()
    negative_group.add_argument(
        "--negative-prompt",
        default=DEFAULT_NEGATIVE_PROMPT,
        help="Negative prompt",
    )
    negative_group.add_argument(
        "--negative-prompt-file",
        type=Path,
        help="Read the negative prompt from a UTF-8 file",
    )
    parser.add_argument("--output", default="outputs/lustify.png", help="Output image path")
    parser.add_argument("--model", type=Path, default=DEFAULT_MODEL, help="Local model directory")
    parser.add_argument(
        "--device",
        default=None,
        help="Torch device, for example cuda:0 or cpu (default: cuda:0 when available)",
    )
    parser.add_argument(
        "--offload",
        choices=("none", "model", "sequential"),
        default="none",
        help="CPU offload mode (default: none)",
    )
    parser.add_argument(
        "--scheduler",
        choices=("dpmpp-2m-sde-karras", "euler-a"),
        default="dpmpp-2m-sde-karras",
    )
    parser.add_argument("--width", type=int, default=1024)
    parser.add_argument("--height", type=int, default=1024)
    parser.add_argument("--steps", type=int, default=30)
    parser.add_argument("--guidance-scale", type=float, default=5.5)
    parser.add_argument("--seed", type=int, default=None)
    parser.add_argument("--num-images", type=int, default=1)
    return parser.parse_args()


def validate_args(args: argparse.Namespace) -> None:
    if not args.model.is_dir():
        raise SystemExit(f"Model directory does not exist: {args.model}")
    for label, path in (
        ("Prompt", args.prompt_file),
        ("Negative prompt", args.negative_prompt_file),
    ):
        if path is not None and not path.is_file():
            raise SystemExit(f"{label} file does not exist: {path}")
    if args.width <= 0 or args.height <= 0 or args.width % 8 or args.height % 8:
        raise SystemExit("--width and --height must be positive multiples of 8")
    if args.steps <= 0:
        raise SystemExit("--steps must be greater than 0")
    if args.num_images <= 0:
        raise SystemExit("--num-images must be greater than 0")


def read_text(direct: str | None, path: Path | None, label: str) -> str:
    text = direct
    if path is not None:
        text = path.read_text(encoding="utf-8").strip()
    if not text:
        raise SystemExit(f"{label} is empty")
    return text


def configure_device(pipe: StableDiffusionXLPipeline, device: str, offload: str) -> None:
    torch_device = torch.device(device)
    if torch_device.type != "cuda":
        pipe.to(torch_device)
        return

    gpu_id = torch_device.index
    if gpu_id is None:
        gpu_id = torch.cuda.current_device()
    if offload == "model":
        pipe.enable_model_cpu_offload(gpu_id=gpu_id)
    elif offload == "sequential":
        pipe.enable_sequential_cpu_offload(gpu_id=gpu_id)
    else:
        pipe.to(torch_device)


def save_images(images: list, output: Path) -> None:
    output.parent.mkdir(parents=True, exist_ok=True)
    if len(images) == 1:
        images[0].save(output)
        print(f"Saved: {output.resolve()}")
        return

    stem, suffix = output.stem, output.suffix or ".png"
    for index, image in enumerate(images, start=1):
        path = output.with_name(f"{stem}_{index:02d}{suffix}")
        image.save(path)
        print(f"Saved: {path.resolve()}")


def main() -> None:
    args = parse_args()
    validate_args(args)
    prompt = read_text(args.prompt, args.prompt_file, "Prompt")
    negative_prompt = read_text(
        args.negative_prompt,
        args.negative_prompt_file,
        "Negative prompt",
    )

    device = args.device or ("cuda:0" if torch.cuda.is_available() else "cpu")
    dtype = torch.float16 if device.startswith("cuda") else torch.float32
    print(f"Loading model from {args.model}")
    print(f"Device: {device} ({dtype}), offload: {args.offload}")

    pipe = StableDiffusionXLPipeline.from_pretrained(
        args.model,
        torch_dtype=dtype,
        use_safetensors=True,
        local_files_only=True,
    )
    if args.scheduler == "dpmpp-2m-sde-karras":
        pipe.scheduler = DPMSolverMultistepScheduler.from_config(
            pipe.scheduler.config,
            algorithm_type="sde-dpmsolver++",
            use_karras_sigmas=True,
        )
    configure_device(pipe, device, args.offload)
    pipe.vae.enable_slicing()
    pipe.set_progress_bar_config(disable=False)

    generator = None
    if args.seed is not None:
        generator = torch.Generator(device="cpu").manual_seed(args.seed)

    result = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        width=args.width,
        height=args.height,
        num_inference_steps=args.steps,
        guidance_scale=args.guidance_scale,
        num_images_per_prompt=args.num_images,
        generator=generator,
    )
    save_images(result.images, Path(args.output))


if __name__ == "__main__":
    main()