File size: 2,799 Bytes
bc5332d
 
 
 
 
 
 
 
 
 
 
 
 
45c6ae9
 
 
 
bc5332d
45c6ae9
 
 
 
 
 
bc5332d
45c6ae9
 
 
 
 
 
 
 
bc5332d
45c6ae9
bc5332d
 
 
ddaa08a
45c6ae9
bc5332d
 
 
 
 
 
 
 
 
 
 
45c6ae9
bc5332d
 
45c6ae9
bc5332d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
import os
from PIL.Image import Image
from diffusers import StableDiffusionPipeline as SDP
from diffusers import AutoPipelineForText2Image as AP
from diffusers import DiffusionPipeline as DP
import torch
import json
import random

DEVICE = "mps" if torch.backends.mps.is_available() else \
        "cuda" if torch.cuda.is_available() else "cpu"

# Model 1
pipe = SDP.from_pretrained("nitrosocke/Ghibli-Diffusion",
                           torch_dtype=torch.float16).to(DEVICE)
ID_PREFIX = "nitrosocke"

# Model 2
# pipe = AP.from_pretrained("black-forest-labs/FLUX.1-dev",
#                           torch_dtype=torch.bfloat16).to(DEVICE)
# pipe.load_lora_weights('openfree/flux-chatgpt-ghibli-lora',
#                        weight_name='flux-chatgpt-ghibli-lora.safetensors')
# ID_PREFIX = "openfree"

# Model 3
# pipe = DP.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0",
#                           torch_dtype=torch.float16, variant="fp16",
#                           use_safetensors=True,).to(DEVICE)
# pipe.load_lora_weights("KappaNeuro/studio-ghibli-style")
# pipe.to(DEVICE)
# pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
# pipe.enable_model_cpu_offload()
# ID_PREFIX = "KappaNeuro"

NUM_IMAGES = 3
out_dir = f"data/{ID_PREFIX}"
os.makedirs(out_dir, exist_ok=True)

with open("metadata.jsonl", "r", encoding="utf-8") as fin, \
    open("ai_entries.jsonl", "w", encoding="utf-8") as fout, \
    open("pairs.jsonl", "a", encoding="utf-8") as pairs:
    for i, line in enumerate(fin):
        sample = json.loads(line)

        if sample["label"] != "real":
            continue

        description = sample["description"]
        real_id: str = sample["id"]
        aigen_id = real_id.replace("real", ID_PREFIX)
        prompt = f"ghibli style, {description}"
        seeds = [random.randrange(2**32) for _ in range(NUM_IMAGES)]
        gens = [torch.Generator(device=DEVICE).manual_seed(s) for s in seeds]

        images: list[Image] = pipe(prompt, num_images_per_prompt=NUM_IMAGES, generator=gens).images
        src_path = sample["image"]
        src_file: str = os.path.basename(src_path)
        file_noext = src_file.split(".")[0]

        for j, image in enumerate(images):
            img_id = f"{aigen_id}-{j}"
            dst_path = os.path.join(out_dir, f"{file_noext}_{j}.jpg")

            fout.write(json.dumps({
                "id": img_id,
                "image": dst_path,
                "label": ID_PREFIX,
                "description": description,
            }) + "\n")
            pairs.write(json.dumps({
                "real_image": src_path,
                "ai_image": dst_path,
                "description": description,
                "seed": seeds[j]
            }) + "\n")
            image.save(dst_path)