style.py

import argparse
import glob
import json
import os

import torch
from accelerate import PartialState
from src_inference.lora_helper import set_single_lora
from src_inference.pipeline import FluxPipeline
from PIL import Image


def clear_cache(transformer):
    for _, attn_processor in transformer.attn_processors.items():
        attn_processor.bank_kv.clear()


class style_processor:
    def __init__(self, flux_path, lora_path, omni_path, device):
        # Initialize model
        self.device = device
        self.base_path = flux_path  # assuming 'flux' is the base path
        self.pipe = FluxPipeline.from_pretrained(
            self.base_path, torch_dtype=torch.bfloat16
        ).to(self.device)
        self.style_prompt = f"{os.path.basename(lora_path).replace('_rank128_bf16.safetensors', '').replace('_', ' ').title()} style, "

        # Load OmniConsistency model
        set_single_lora(
            self.pipe.transformer,
            omni_path,
            lora_weights=[1],
            cond_size=512,
        )

        # Load external LoRA
        self.pipe.unload_lora_weights()
        self.pipe.load_lora_weights(lora_path, weight_name="lora_name.safetensors")

    def process(self, image_path, prompt):
        if isinstance(image_path, str):
            spatial_image = [Image.open(image_path).convert("RGB")]
        elif isinstance(image_path, Image.Image):
            spatial_image = [image_path]
        else:
            raise ValueError(f"Invalid image type: {type(image_path)}")

        subject_images = []

        width, height = spatial_image[0].size

        image = self.pipe(
            prompt,
            height=height,
            width=width,
            guidance_scale=3.5,
            num_inference_steps=25,
            max_sequence_length=512,
            generator=torch.Generator("cpu").manual_seed(5),
            spatial_images=spatial_image,
            subject_images=subject_images,
            cond_size=512,
        ).images[0]

        # Clear cache after generation
        clear_cache(self.pipe.transformer)

        return image


def get_images_from_path(path):
    if os.path.isdir(path):
        return glob.glob(os.path.join(path, "*.jpg")) + glob.glob(
            os.path.join(path, "*.png")
        )
    elif os.path.isfile(path) and (path.endswith(".jpg") or path.endswith(".png")):
        return [path]
    else:
        return []


def parse_args():
    parser = argparse.ArgumentParser(description="Style processor")
    parser.add_argument("--flux_path", type=str, required=True)
    parser.add_argument("--lora_paths", type=str, required=True, nargs="+")
    parser.add_argument("--omni_path", type=str, required=True)
    parser.add_argument("--output_dir", type=str, required=True)
    parser.add_argument("--prompt_dir", type=str, required=True)
    parser.add_argument("--images_path", type=str, required=True)
    return parser.parse_args()


if __name__ == "__main__":
    args = parse_args()
    flux_path = args.flux_path
    lora_paths = args.lora_paths
    omni_path = args.omni_path
    output_dir = args.output_dir
    prompt_dir = args.prompt_dir
    images_path = args.images_path

    distributed_state = PartialState()

    device = distributed_state.device
    rank = int(str(device).split(":")[1])
    lora = lora_paths[rank]

    output_lora_path = os.path.join(output_dir, os.path.basename(lora))
    os.makedirs(output_lora_path, exist_ok=True)

    processor = style_processor(flux_path, lora, omni_path, device)

    images_path = get_images_from_path(images_path)
    for image_path in images_path:
        image_output_path = os.path.join(output_lora_path, os.path.basename(image_path))
        if os.path.exists(image_output_path):
            print(f"File {image_output_path} already exists, skipping.")
            continue

        try:
            with open(
                os.path.join(prompt_dir, os.path.basename(image_path) + ".json")
            ) as f:
                prompt = json.load(f)["caption"]
            output = processor.process(image_path, processor.style_prompt + prompt)
            output.save(image_output_path)
        except Exception as e:
            print(f"Error processing {image_path}: {e}")