Create app.py

app.py

@@ -0,0 +1,160 @@
+import spaces
+import gradio as gr
+import numpy as np
+from PIL import Image
+import random
+from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
+import torch
+from transformers import pipeline as transformers_pipeline
+import re
+
+# Device selection for image generation (GPU if available)
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+# Stable Diffusion XL pipeline
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    "votepurchase/waiNSFWIllustrious_v120",
+    torch_dtype=torch.float16,
+    variant="fp16",
+    use_safetensors=True,
+)
+pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
+pipe.to(device)
+
+# Force modules to fp16 for memory efficiency
+pipe.text_encoder.to(torch.float16)
+pipe.text_encoder_2.to(torch.float16)
+pipe.vae.to(torch.float16)
+pipe.unet.to(torch.float16)
+
+# Korean → English translator (CPU only)
+translator = transformers_pipeline(
+    "translation",
+    model="Helsinki-NLP/opus-mt-ko-en",
+    device=-1,  # -1 forces CPU
+)
+
+MAX_SEED = np.iinfo(np.int32).max
+MAX_IMAGE_SIZE = 1216
+korean_regex = re.compile("[\uac00-\ud7af]+")
+
+def maybe_translate(text: str) -> str:
+    """Translate Korean text to English if Korean characters are detected."""
+    if korean_regex.search(text):
+        translation = translator(text, max_length=256, clean_up_tokenization_spaces=True)
+        return translation[0]["translation_text"]
+    return text
+
+@spaces.GPU
+def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
+    prompt = maybe_translate(prompt)
+    negative_prompt = maybe_translate(negative_prompt)
+
+    if len(prompt.split()) > 60:
+        print("Warning: Prompt may be too long and will be truncated by the model")
+
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
+
+    generator = torch.Generator(device=device).manual_seed(seed)
+
+    try:
+        output_image = pipe(
+            prompt=prompt,
+            negative_prompt=negative_prompt,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            width=width,
+            height=height,
+            generator=generator,
+        ).images[0]
+        return output_image
+    except RuntimeError as e:
+        print(f"Error during generation: {e}")
+        error_img = Image.new("RGB", (width, height), color=(0, 0, 0))
+        return error_img
+
+# Custom styling
+css = """
+body {background: #0f0f0f; color: #fafafa; font-family: 'Noto Sans', sans-serif;}
+#col-container {margin: 0 auto; max-width: 640px;}
+.gr-button {background: #2563eb; color: #ffffff; border-radius: 8px;}
+#prompt-box textarea {font-size: 1.1rem; height: 3rem;}
+"""
+
+with gr.Blocks(css=css, theme=gr.themes.Soft()) as demo:
+    gr.Markdown(
+        """
+        ## 🖌️ Stable Diffusion XL Playground  
+        Generate high quality illustrations with a single prompt.  
+        **Tip:** Write in Korean or English. Korean will be translated automatically.
+        """
+    )
+
+    with gr.Column(elem_id="col-container"):
+        with gr.Row():
+            prompt = gr.Text(
+                label="Prompt",
+                elem_id="prompt-box",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt (60 words max)",
+            )
+            run_button = gr.Button("Generate", scale=0)
+
+        result = gr.Image(label="", show_label=False)
+
+        examples = gr.Examples(
+            examples=[
+                ["어두운 재즈 바에서 담배 연기를 내뿜는 미스터리한 팜파탈, 성인용 애니메이션 스타일"],
+                ["노출이 강조된 드레스를 입은 고딕 뱀파이어 여왕, 드라마틱 조명, 성인 애니 아트"],
+                ["은은한 조명의 온천에서 두 연인이 마주 서 있는 관능적 장면, 성인용 애니메이션"],
+                ["네온이 빛나는 사이버펑크 클럽 무대에서 도발적인 의상을 입은 댄서, 성인 애니 스타일"],
+                ["달빛 아래 요염한 마법사가 주문을 외우는 판타지 장면, 성인용 애니 일러스트"],
+            ],
+            inputs=[prompt],
+        )
+
+        with gr.Accordion("Advanced Settings", open=False):
+            negative_prompt = gr.Text(
+                label="Negative prompt",
+                max_lines=1,
+                placeholder="Enter a negative prompt",
+                value="nsfw, low quality, watermark, signature",
+            )
+
+            seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
+            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+
+            with gr.Row():
+                width = gr.Slider(
+                    label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024
+                )
+                height = gr.Slider(
+                    label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=1024
+                )
+
+            with gr.Row():
+                guidance_scale = gr.Slider(
+                    label="Guidance scale", minimum=0.0, maximum=20.0, step=0.1, value=7
+                )
+                num_inference_steps = gr.Slider(
+                    label="Inference steps", minimum=1, maximum=28, step=1, value=28
+                )
+
+    run_button.click(
+        fn=infer,
+        inputs=[
+            prompt,
+            negative_prompt,
+            seed,
+            randomize_seed,
+            width,
+            height,
+            guidance_scale,
+            num_inference_steps,
+        ],
+        outputs=[result],
+    )
+
+demo.queue().launch()