app.py

import random
import gradio as gr
import numpy as np
import torch
import spaces
from diffusers import FluxPipeline
from PIL import Image
from diffusers.utils import export_to_gif
from transformers import pipeline

HEIGHT = 256
WIDTH = 1024
MAX_SEED = np.iinfo(np.int32).max

device = "cuda" if torch.cuda.is_available() else "cpu"
pipe = FluxPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev",
    torch_dtype=torch.bfloat16
).to(device)

translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")

def split_image(input_image, num_splits=4):
    output_images = []
    for i in range(num_splits):
        left = i * 256
        right = (i + 1) * 256
        box = (left, 0, right, 256)
        output_images.append(input_image.crop(box))
    return output_images

def translate_to_english(text):
    return translator(text)[0]['translation_text']

@spaces.GPU(duration=190)
def predict(prompt, seed=42, randomize_seed=False, guidance_scale=5.0, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
    if any('\u3131' <= char <= '\u318E' or '\uAC00' <= char <= '\uD7A3' for char in prompt):
        prompt = translate_to_english(prompt)
    
    prompt_template = f"""
    A side by side 4 frame image showing consecutive stills from a looped gif moving from left to right. The gif is of {prompt}.
    """
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    image = pipe(
        prompt=prompt_template,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        num_images_per_prompt=1,
        generator=torch.Generator("cpu").manual_seed(seed),
        height=HEIGHT,
        width=WIDTH
    ).images[0]
    
    return export_to_gif(split_image(image, 4), "flux.gif", fps=4), image, seed

css = """
footer { visibility: hidden;}
"""

examples = [
    "고양이가 공중에서 발을 흔드는 모습",
    "팬더가 엉덩이를 좌우로 흔드는 모습",
    "꽃이 피어나는 과정"
]

with gr.Blocks(theme="Nymbo/Nymbo_Theme", css=css) as demo:
    with gr.Column(elem_id="col-container"):
        with gr.Row():
            prompt = gr.Text(label="프롬프트", show_label=False, max_lines=1, placeholder="프롬프트를 입력하세요")
            submit = gr.Button("제출", scale=0)
        output = gr.Image(label="GIF", show_label=False)
        output_stills = gr.Image(label="스틸 이미지", show_label=False, elem_id="stills")
        
        with gr.Accordion("고급 설정", open=False):
            seed = gr.Slider(
                label="시드",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )
            randomize_seed = gr.Checkbox(label="시드 무작위화", value=True)
            with gr.Row():
                guidance_scale = gr.Slider(
                    label="가이던스 스케일",
                    minimum=1,
                    maximum=15,
                    step=0.1,
                    value=3.5,
                )
                num_inference_steps = gr.Slider(
                    label="추론 단계 수",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )
        
        gr.Examples(
            examples=examples,
            fn=predict,
            inputs=[prompt],
            outputs=[output, output_stills, seed],
            cache_examples="lazy"
        )
        
        gr.on(
            triggers=[submit.click, prompt.submit],
            fn=predict,
            inputs=[prompt, seed, randomize_seed, guidance_scale, num_inference_steps],
            outputs=[output, output_stills, seed]
        )

demo.launch()