Initial commit: Stable Diffusion image generator

.gitignore

@@ -0,0 +1,12 @@
+__pycache__/
+*.py[cod]
+*.so
+.Python
+*.egg-info/
+flagged/
+gradio_cached_examples/
+.DS_Store
+*.png
+*.jpg
+*.jpeg
+

README.md

@@ -1,13 +1,82 @@
----
-title: Spac3imag3
-emoji: 🏆
-colorFrom: pink
-colorTo: red
+title: Stable Diffusion Generator
+emoji: 🎨
+colorFrom: blue
+colorTo: cyan
 sdk: gradio
-sdk_version: 5.49.1
+sdk_version: 4.19.2
 app_file: app.py
 pinned: false
 license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🎨 Stable Diffusion Image Generator
+
+Generate stunning AI art from text descriptions using Stable Diffusion 2.1!
+
+## 🚀 Features
+
+- 💬 **Text-to-Image**: Create images from detailed descriptions
+- 🎨 **Customizable**: Control size, quality, and style
+- 🎲 **Reproducible**: Use seeds for consistent results
+- ⚡ **Multiple Schedulers**: DPM++ 2M and Euler Ancestral
+- 🖼️ **Batch Generation**: Create multiple variations at once
+
+## 🎯 How to Use
+
+1. **Write a Prompt**: Describe the image you want in detail
+2. **Add Negative Prompt**: Specify what to avoid
+3. **Adjust Settings**: Size, steps, guidance scale
+4. **Generate**: Click the button and wait
+5. **Download**: Save your AI-generated artwork!
+
+## 💡 Prompt Tips
+
+### Good Prompt Structure
+`[Subject] + [Style] + [Lighting] + [Details] + [Quality]`
+
+### Examples
+- "a serene japanese garden, cherry blossoms, koi pond, soft lighting, 4k"
+- "portrait of a cyberpunk character, neon lights, detailed, digital art"
+- "fantasy landscape with mountains, dramatic sunset, epic, highly detailed"
+
+### Style Keywords
+- photorealistic, digital art, oil painting, watercolor, anime
+- cinematic, ethereal, vibrant, dramatic, soft
+
+### Quality Modifiers
+- highly detailed, 4k, 8k, sharp focus, intricate
+- trending on artstation, masterpiece
+
+## 🔧 Settings Guide
+
+| Parameter | Range | Recommended | Effect |
+|-----------|-------|-------------|--------|
+| **Steps** | 10-50 | 20-30 | Quality (more = better but slower) |
+| **Guidance** | 1-20 | 7-10 | Prompt adherence |
+| **Size** | 256-1024 | 512x512 | Image resolution |
+| **Seed** | Any number | -1 (random) | Reproducibility |
+
+## 📊 Performance
+
+| Hardware | Resolution | Steps | Time |
+|----------|-----------|-------|------|
+| CPU Free | 512x512 | 20 | ~3-5 min |
+| T4 GPU | 512x512 | 25 | ~15-20 sec |
+| T4 GPU | 768x768 | 30 | ~30-40 sec |
+
+## 🎨 Model Info
+
+- **Base**: Stable Diffusion 2.1
+- **Resolution**: Optimized for 768px
+- **Training**: LAION-5B dataset
+- **License**: CreativeML Open RAIL++-M
+
+## 📚 Resources
+
+- [Stable Diffusion Paper](https://arxiv.org/abs/2112.10752)
+- [Model Card](https://huggingface.co/stabilityai/stable-diffusion-2-1)
+- [Diffusers Docs](https://huggingface.co/docs/diffusers)
+
+---
+
+**Made with ❤️ using HuggingFace Diffusers**

app.py

@@ -0,0 +1,415 @@
+import gradio as gr
+import torch
+from diffusers import (
+    StableDiffusionPipeline,
+    DPMSolverMultistepScheduler,
+    EulerAncestralDiscreteScheduler
+)
+from PIL import Image
+import random
+
+print("🎨 Initializing Stable Diffusion pipeline...")
+
+# Configurazione
+MODEL_ID = "stabilityai/stable-diffusion-2-1"
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+USE_SAFETENSORS = True
+
+# Carica pipeline
+print(f"📦 Loading model: {MODEL_ID}")
+
+pipe = StableDiffusionPipeline.from_pretrained(
+    MODEL_ID,
+    torch_dtype=torch.float16 if DEVICE == "cuda" else torch.float32,
+    use_safetensors=USE_SAFETENSORS,
+    safety_checker=None,  # Disabilita per velocità (opzionale)
+)
+
+# Scheduler ottimizzato
+pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
+
+pipe.to(DEVICE)
+
+# Ottimizzazioni
+if DEVICE == "cuda":
+    pipe.enable_model_cpu_offload()
+    pipe.enable_vae_slicing()
+    print("✅ GPU optimizations enabled")
+
+print(f"✅ Pipeline loaded on {DEVICE}")
+
+
+def generate_image(
+    prompt,
+    negative_prompt="",
+    width=512,
+    height=512,
+    num_inference_steps=25,
+    guidance_scale=7.5,
+    num_images=1,
+    seed=-1,
+    scheduler_type="DPM++ 2M",
+    progress=gr.Progress()
+):
+    """
+    Genera immagini da prompt testuale
+    
+    Args:
+        prompt: Descrizione dell'immagine da generare
+        negative_prompt: Cosa evitare
+        width: Larghezza immagine (multiplo di 8)
+        height: Altezza immagine (multiplo di 8)
+        num_inference_steps: Step di qualità (15-50)
+        guidance_scale: Aderenza al prompt (5-15)
+        num_images: Numero di immagini da generare (1-4)
+        seed: Random seed (-1 per random)
+        scheduler_type: Tipo di scheduler
+        progress: Progress tracker
+    """
+    
+    if not prompt or len(prompt.strip()) == 0:
+        return None, "❌ Inserisci un prompt!"
+    
+    try:
+        progress(0, desc="🎨 Initializing generation...")
+        
+        # Imposta seed
+        if seed == -1:
+            seed = random.randint(0, 2147483647)
+        
+        generator = torch.Generator(device=DEVICE).manual_seed(seed)
+        
+        print(f"📝 Prompt: {prompt}")
+        print(f"🎲 Seed: {seed}")
+        print(f"📐 Size: {width}x{height}")
+        print(f"🎨 Steps: {num_inference_steps}")
+        
+        # Cambia scheduler se richiesto
+        if scheduler_type == "Euler a":
+            pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(
+                pipe.scheduler.config
+            )
+        elif scheduler_type == "DPM++ 2M":
+            pipe.scheduler = DPMSolverMultistepScheduler.from_config(
+                pipe.scheduler.config
+            )
+        
+        progress(0.2, desc="🖼️ Generating image...")
+        
+        # Genera immagini
+        with torch.no_grad():
+            result = pipe(
+                prompt=prompt,
+                negative_prompt=negative_prompt if negative_prompt else None,
+                width=width,
+                height=height,
+                num_inference_steps=num_inference_steps,
+                guidance_scale=guidance_scale,
+                num_images_per_prompt=num_images,
+                generator=generator,
+            )
+        
+        images = result.images
+        
+        progress(1.0, desc="✅ Complete!")
+        
+        # Info
+        info = f"""
+        ✅ **Immagine generata con successo!**
+        
+        📊 **Dettagli:**
+        - Prompt: "{prompt}"
+        - Negative: "{negative_prompt if negative_prompt else 'None'}"
+        - Risoluzione: {width}x{height}
+        - Steps: {num_inference_steps}
+        - Guidance Scale: {guidance_scale}
+        - Seed: {seed}
+        - Scheduler: {scheduler_type}
+        - Device: {DEVICE.upper()}
+        - Immagini generate: {len(images)}
+        
+        💡 **Tip:** Salva il seed per ricreare immagini simili!
+        """
+        
+        # Restituisci prima immagine + gallery
+        return images[0], images if len(images) > 1 else None, info
+        
+    except Exception as e:
+        error_msg = f"""
+        ❌ **Errore durante la generazione:**
+        
+        {str(e)}
+        
+        💡 **Possibili soluzioni:**
+        - Riduci risoluzione (512x512 consigliato)
+        - Riduci inference steps (20-25)
+        - Semplifica il prompt
+        - Verifica che width e height siano multipli di 8
+        """
+        print(f"Error: {e}")
+        import traceback
+        traceback.print_exc()
+        return None, None, error_msg
+
+
+# Esempi predefiniti
+EXAMPLES = [
+    [
+        "a beautiful landscape with mountains and a lake at sunset, highly detailed, 4k",
+        "blurry, low quality, distorted, ugly",
+        512, 512, 25, 7.5, 1, 42, "DPM++ 2M"
+    ],
+    [
+        "portrait of a cute cat wearing a wizard hat, digital art, detailed fur",
+        "low quality, blurry",
+        512, 512, 30, 8.0, 1, 123, "DPM++ 2M"
+    ],
+    [
+        "futuristic city with flying cars, neon lights, cyberpunk style, detailed",
+        "blurry, low quality",
+        768, 512, 25, 7.5, 1, 456, "DPM++ 2M"
+    ],
+    [
+        "medieval castle on a hill, dramatic lighting, fantasy art, intricate details",
+        "modern, contemporary",
+        512, 768, 30, 7.5, 1, 789, "Euler a"
+    ],
+]
+
+
+# Interfaccia Gradio
+with gr.Blocks(
+    title="🎨 Stable Diffusion Generator",
+    theme=gr.themes.Soft(
+        primary_hue="blue",
+        secondary_hue="cyan"
+    ),
+    css="""
+    .gradio-container {max-width: 1400px !important}
+    """
+) as demo:
+    
+    gr.Markdown("""
+    # 🎨 Stable Diffusion Image Generator
+    ### Create Stunning AI Art from Text Descriptions
+    
+    Powered by **Stable Diffusion 2.1** - State-of-the-art text-to-image generation
+    
+    💡 **Tips for better results:**
+    - Be specific and descriptive
+    - Mention style, lighting, and details
+    - Use negative prompts to avoid unwanted elements
+    - Experiment with different seeds and settings
+    """)
+    
+    with gr.Row():
+        # Colonna sinistra - Input
+        with gr.Column(scale=1):
+            prompt_input = gr.Textbox(
+                label="✨ Prompt (Describe what you want to create)",
+                placeholder="Example: a serene japanese garden with cherry blossoms, koi pond, soft lighting, highly detailed, 4k",
+                lines=4,
+                value="a beautiful landscape with mountains and a lake at sunset, highly detailed, 4k"
+            )
+            
+            negative_prompt_input = gr.Textbox(
+                label="🚫 Negative Prompt (What to avoid)",
+                placeholder="Example: blurry, low quality, distorted, ugly, deformed",
+                lines=2,
+                value="blurry, low quality, distorted, ugly"
+            )
+            
+            with gr.Row():
+                width = gr.Slider(
+                    minimum=256,
+                    maximum=1024,
+                    value=512,
+                    step=64,
+                    label="📐 Width",
+                    info="Must be multiple of 64"
+                )
+                
+                height = gr.Slider(
+                    minimum=256,
+                    maximum=1024,
+                    value=512,
+                    step=64,
+                    label="📐 Height",
+                    info="Must be multiple of 64"
+                )
+            
+            with gr.Accordion("⚙️ Advanced Settings", open=False):
+                num_inference_steps = gr.Slider(
+                    minimum=10,
+                    maximum=50,
+                    value=25,
+                    step=5,
+                    label="🎨 Inference Steps",
+                    info="More = better quality but slower"
+                )
+                
+                guidance_scale = gr.Slider(
+                    minimum=1.0,
+                    maximum=20.0,
+                    value=7.5,
+                    step=0.5,
+                    label="🎯 Guidance Scale",
+                    info="How closely to follow the prompt (7-10 recommended)"
+                )
+                
+                num_images = gr.Slider(
+                    minimum=1,
+                    maximum=4,
+                    value=1,
+                    step=1,
+                    label="🖼️ Number of Images",
+                    info="Generate multiple variations"
+                )
+                
+                scheduler_type = gr.Dropdown(
+                    choices=["DPM++ 2M", "Euler a"],
+                    value="DPM++ 2M",
+                    label="🔧 Scheduler",
+                    info="Different sampling methods"
+                )
+                
+                seed = gr.Number(
+                    value=-1,
+                    label="🎲 Seed (-1 for random)",
+                    info="Use same seed for consistent results",
+                    precision=0
+                )
+            
+            generate_btn = gr.Button(
+                "🎨 Generate Image",
+                variant="primary",
+                size="lg"
+            )
+            
+            gr.Markdown("""
+            ### 📊 Performance Guide
+            
+            **CPU (Free tier):**
+            - Resolution: 512x512
+            - Steps: 15-20
+            - Time: ~2-5 min
+            
+            **GPU T4 ($0.60/h):**
+            - Resolution: 768x768
+            - Steps: 25-35
+            - Time: ~10-30 sec
+            """)
+        
+        # Colonna destra - Output
+        with gr.Column(scale=1):
+            image_output = gr.Image(
+                label="🖼️ Generated Image",
+                type="pil",
+                height=512
+            )
+            
+            gallery_output = gr.Gallery(
+                label="🎨 Image Variations",
+                columns=2,
+                rows=2,
+                height=400,
+                visible=False
+            )
+            
+            info_output = gr.Markdown(
+                value="👆 Write a prompt and click 'Generate' to create your image!",
+                label="ℹ️ Generation Info"
+            )
+    
+    # Sezione esempi
+    gr.Markdown("### 🎨 Example Prompts - Click to try")
+    
+    gr.Examples(
+        examples=EXAMPLES,
+        inputs=[
+            prompt_input,
+            negative_prompt_input,
+            width,
+            height,
+            num_inference_steps,
+            guidance_scale,
+            num_images,
+            seed,
+            scheduler_type
+        ],
+        outputs=[image_output, gallery_output, info_output],
+        fn=generate_image,
+        cache_examples=False,
+    )
+    
+    # Event handler
+    generate_btn.click(
+        fn=generate_image,
+        inputs=[
+            prompt_input,
+            negative_prompt_input,
+            width,
+            height,
+            num_inference_steps,
+            guidance_scale,
+            num_images,
+            seed,
+            scheduler_type
+        ],
+        outputs=[image_output, gallery_output, info_output],
+    )
+    
+    # Footer
+    gr.Markdown("""
+    ---
+    ### 📚 Prompt Engineering Guide
+    
+    **Structure:** `[Subject] + [Style] + [Lighting] + [Details] + [Quality]`
+    
+    **Good Prompt Examples:**
+    - "a serene japanese garden with cherry blossoms, koi pond, soft golden hour lighting, highly detailed, 4k, photorealistic"
+    - "portrait of an astronaut floating in space, cinematic lighting, digital art, trending on artstation"
+    - "fantasy castle on a floating island, dramatic storm clouds, epic scale, concept art, octane render"
+    
+    **Style Keywords:**
+    - photorealistic, digital art, oil painting, watercolor, anime, concept art
+    - cinematic, dramatic, ethereal, vibrant, muted, pastel
+    
+    **Quality Modifiers:**
+    - highly detailed, 4k, 8k, ultra detailed, intricate, sharp focus
+    - trending on artstation, award winning, masterpiece
+    
+    **Common Negative Prompts:**
+    - blurry, low quality, distorted, ugly, deformed, duplicate
+    - bad anatomy, poorly drawn, amateur, watermark, signature
+    
+    ---
+    
+    ### 🔧 Technical Details
+    
+    - **Model**: Stable Diffusion 2.1 (768px base model)
+    - **Scheduler**: DPM++ 2M Karras / Euler Ancestral
+    - **Device**: {device}
+    - **VRAM**: Optimized with CPU offload and VAE slicing
+    
+    ### 💡 Tips
+    
+    - **Square images** (512x512) are fastest
+    - **Portrait** (512x768) or **Landscape** (768x512) for specific ratios
+    - Start with **guidance scale 7-8**, adjust if needed
+    - Use **20-25 steps** for good quality
+    - Save your **seed** to recreate variations
+    
+    ---
+    
+    **Made with ❤️ using HuggingFace Diffusers & Stable Diffusion**
+    """.replace("{device}", DEVICE.upper()))
+
+# Launch
+if __name__ == "__main__":
+    demo.queue(max_size=20)
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=False
+    )

requirements.txt

@@ -0,0 +1,10 @@
+diffusers==0.27.2
+transformers==4.38.1
+accelerate==0.27.2
+torch==2.2.0
+torchvision==0.17.0
+gradio==4.19.2
+pillow==10.2.0
+numpy==1.26.4
+safetensors==0.4.2
+huggingface-hub==0.21.4