import gradio as gr from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline from diffusers import StableDiffusionInpaintPipeline import torch from PIL import Image, ImageDraw, ImageFont import time import os import tempfile import random # === OPTIMIERTE EINSTELLUNGEN === device = "cuda" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if device == "cuda" else torch.float32 IMG_SIZE = 512 print(f"Running on: {device}") # === TEXT INTEGRATION IMPORT === from text_integration import ( add_text_to_image, create_text_integration_section_t2i, create_text_integration_section_i2i, capture_click, update_text_preview_i2i, update_text_preview_t2i ) # === GESICHTSMASKEN-FUNKTIONEN === def create_face_mask(image, bbox_coords, face_preserve): """Erzeugt eine Gesichtsmaske - WEIßE Bereiche werden VERÄNDERT, SCHWARZE BLEIBEN""" mask = Image.new("L", image.size, 0) if bbox_coords and all(coord is not None for coord in bbox_coords): x1, y1, x2, y2 = bbox_coords draw = ImageDraw.Draw(mask) if face_preserve: draw.rectangle([0, 0, image.size[0], image.size[1]], fill=255) draw.rectangle([x1, y1, x2, y2], fill=0) print("Gesicht wird GESCHÜTZT - Umgebung wird verändert") else: draw.rectangle([x1, y1, x2, y2], fill=255) print("Nur Gesicht wird verändert - Umgebung bleibt erhalten") return mask def auto_detect_face_area(image): """Optimierten Vorschlag für Gesichtsbereich ohne externe Bibliotheken""" width, height = image.size face_size = min(width, height) * 0.4 x1 = (width - face_size) / 2 y1 = (height - face_size) / 4 x2 = x1 + face_size y2 = y1 + face_size * 1.2 x1, y1 = max(0, int(x1)), max(0, int(y1)) x2, y2 = min(width, int(x2)), min(height, int(y2)) print(f"Geschätzte Gesichtskoordinaten: [{x1}, {y1}, {x2}, {y2}]") return [x1, y1, x2, y2] # === PIPELINES === pipe_txt2img = None pipe_img2img = None def load_txt2img(): global pipe_txt2img if pipe_txt2img is None: print("Loading Text-to-Image model...") pipe_txt2img = StableDiffusionPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", torch_dtype=torch_dtype, use_safetensors=True, safety_checker=None, requires_safety_checker=False, ).to(device) from diffusers import DPMSolverMultistepScheduler pipe_txt2img.scheduler = DPMSolverMultistepScheduler.from_config(pipe_txt2img.scheduler.config) pipe_txt2img.enable_attention_slicing() return pipe_txt2img def load_img2img(): global pipe_img2img if pipe_img2img is None: print("Loading Inpainting model...") try: pipe_img2img = StableDiffusionInpaintPipeline.from_pretrained( "runwayml/stable-diffusion-inpainting", torch_dtype=torch_dtype, allow_pickle=False, safety_checker=None, ).to(device) except Exception as e: print(f"Fehler beim Laden des Modells: {e}") raise from diffusers import DPMSolverMultistepScheduler pipe_img2img.scheduler = DPMSolverMultistepScheduler.from_config( pipe_img2img.scheduler.config, algorithm_type="sde-dpmsolver++", use_karras_sigmas=True, timestep_spacing="trailing" ) pipe_img2img.enable_attention_slicing() pipe_img2img.enable_vae_tiling() pipe_img2img.vae_slicing = True return pipe_img2img # === CALLBACK-FUNKTIONEN === class TextToImageProgressCallback: def __init__(self, progress, total_steps): self.progress = progress self.total_steps = total_steps self.current_step = 0 def __call__(self, pipe, step, timestep, callback_kwargs): self.current_step = step + 1 progress_percent = (step / self.total_steps) * 100 self.progress(progress_percent / 100, desc="Generierung läuft - CPU benötigt bis zu 20 Minuten!") return callback_kwargs class ImageToImageProgressCallback: def __init__(self, progress, total_steps, strength): self.progress = progress self.total_steps = total_steps self.current_step = 0 self.strength = strength self.actual_total_steps = None def __call__(self, pipe, step, timestep, callback_kwargs): self.current_step = step + 1 if self.actual_total_steps is None: if self.strength < 1.0: self.actual_total_steps = int(self.total_steps * self.strength) else: self.actual_total_steps = self.total_steps print(f"🎯 INTERNE STEP-AUSGABE: Strength {self.strength} → {self.actual_total_steps} tatsächliche Denoising-Schritte") progress_percent = (step / self.actual_total_steps) * 100 self.progress(progress_percent / 100, desc="Generierung läuft - CPU benötigt bis zu 20 Minuten!") return callback_kwargs # === VORSCHAU-FUNKTIONEN === def create_preview_image(image, bbox_coords, face_preserve, mode_color): """Erstellt eine Vorschau mit farbigem Rahmen basierend auf dem Modus""" if image is None: return None preview = image.copy() draw = ImageDraw.Draw(preview) if mode_color == "red": border_color = (255, 0, 0, 180) mode_text = "NUR BILDELEMENT VERÄNDERN" else: border_color = (0, 255, 0, 180) mode_text = "BILDELEMENT BEIBEHALTEN" border_width = 8 draw.rectangle([0, 0, preview.width-1, preview.height-1], outline=border_color, width=border_width) if bbox_coords and all(coord is not None for coord in bbox_coords): x1, y1, x2, y2 = bbox_coords box_color = (255, 255, 0, 200) draw.rectangle([x1, y1, x2, y2], outline=box_color, width=3) text_color = (255, 255, 255) bg_color = (0, 0, 0, 160) text_bbox = draw.textbbox((x1, y1 - 25), mode_text) draw.rectangle([text_bbox[0]-5, text_bbox[1]-2, text_bbox[2]+5, text_bbox[3]+2], fill=bg_color) draw.text((x1, y1 - 25), mode_text, fill=text_color) return preview def update_live_preview(image, bbox_x1, bbox_y1, bbox_x2, bbox_y2, face_preserve): """Aktualisiert die Live-Vorschau bei Koordinaten-Änderungen""" if image is None: return None bbox_coords = [bbox_x1, bbox_y1, bbox_x2, bbox_y2] mode_color = "green" if face_preserve else "red" return create_preview_image(image, bbox_coords, face_preserve, mode_color) def process_image_upload(image): """Verarbeitet Bild-Upload und gibt Bild + Koordinaten zurück""" if image is None: return None, None, None, None, None bbox = auto_detect_face_area(image) bbox_x1, bbox_y1, bbox_x2, bbox_y2 = bbox preview = create_preview_image(image, bbox, True, "green") return preview, bbox_x1, bbox_y1, bbox_x2, bbox_y2 # === HAUPTPROZESSE === def text_to_image(prompt, steps, guidance_scale, progress=gr.Progress()): try: if not prompt or not prompt.strip(): return None print(f"Starting generation for: {prompt}") start_time = time.time() progress(0, desc="Generierung läuft - CPU benötigt bis zu 20 Minuten!") pipe = load_txt2img() seed = random.randint(0, 2**32 - 1) generator = torch.Generator(device=device).manual_seed(seed) print(f"Using seed: {seed}") callback = TextToImageProgressCallback(progress, steps) image = pipe( prompt=prompt, height=IMG_SIZE, width=IMG_SIZE, num_inference_steps=int(steps), guidance_scale=guidance_scale, generator=generator, callback_on_step_end=callback, callback_on_step_end_tensor_inputs=[], ).images[0] end_time = time.time() print(f"Bild generiert in {end_time - start_time:.2f} Sekunden") return image except Exception as e: print(f"Fehler: {e}") import traceback traceback.print_exc() return None def img_to_image(image, prompt, neg_prompt, strength, steps, guidance_scale, face_preserve, bbox_x1, bbox_y1, bbox_x2, bbox_y2, progress=gr.Progress()): try: if image is None: return None print(f"Img2Img Start → Strength: {strength}, Steps: {steps}, Guidance: {guidance_scale}") start_time = time.time() progress(0, desc="Generierung läuft - CPU benötigt bis zu 20 Minuten!") pipe = load_img2img() img_resized = image.convert("RGB").resize((IMG_SIZE, IMG_SIZE)) adj_strength = min(0.85, strength * 1.3) adj_guidance = min(guidance_scale, 12.0) seed = random.randint(0, 2**32 - 1) generator = torch.Generator(device=device).manual_seed(seed) print(f"Using seed: {seed}") mask = None bbox_coords = None if bbox_x1 is not None and bbox_y1 is not None and bbox_x2 is not None and bbox_y2 is not None: orig_width, orig_height = image.size scale_x = IMG_SIZE / orig_width scale_y = IMG_SIZE / orig_height scaled_coords = [ int(bbox_x1 * scale_x), int(bbox_y1 * scale_y), int(bbox_x2 * scale_x), int(bbox_y2 * scale_y) ] bbox_coords = scaled_coords if bbox_coords: mask = create_face_mask(img_resized, bbox_coords, face_preserve) callback = ImageToImageProgressCallback(progress, int(steps), adj_strength) result = pipe( prompt=prompt, negative_prompt=neg_prompt, image=img_resized, mask_image=mask, strength=adj_strength, num_inference_steps=int(steps), guidance_scale=adj_guidance, generator=generator, callback_on_step_end=callback, callback_on_step_end_tensor_inputs=[], ) end_time = time.time() print(f"Bild transformiert in {end_time - start_time:.2f} Sekunden") generated_image = result.images[0] return generated_image except Exception as e: print(f"Fehler: {e}") import traceback traceback.print_exc() return None # === TEXT INTEGRATION HANDLER === def handle_text_integration_i2i(original_image, generated_image, text, text_x, text_y, target_selector): """Verwaltet Text-Integration für Bild-zu-Bild basierend auf Auswahl""" if target_selector == "Originalbild": target_image = original_image else: # "Generiertes Bild" target_image = generated_image result = add_text_to_image(target_image, text, text_x, text_y) # Rückgabe: Original bleibt unverändert, Text-Bild kommt in Download-Bereich return original_image, result def handle_text_integration_t2i(generated_image, text, text_x, text_y): """Verwaltet Text-Integration für Text-zu-Bild""" result = add_text_to_image(generated_image, text, text_x, text_y) return result def main_ui(): with gr.Blocks( title="AI Image Generator", theme=gr.themes.Base(), css=""" .info-box { background-color: #f8f4f0; padding: 15px; border-radius: 8px; border-left: 4px solid #8B7355; margin: 20px 0; } .text-integration-section { background: #e8f5e8; padding: 15px; border-radius: 8px; margin: 15px 0; border-left: 4px solid #4caf50; } """ ) as demo: # --- Info-Bereich --- gr.Markdown("# AI Image Generator") with gr.Row(): with gr.Column(scale=1): pass with gr.Column(scale=1, min_width=300): start_btn = gr.Button("Weiter zur Anwendung", variant="primary", size="lg") with gr.Column(scale=1): pass # --- Hauptanwendungsbereich --- with gr.Column(visible=False) as content_area: # === TAB: TEXT ZU BILD === with gr.Tab("Text zu Bild"): gr.Markdown("**Beschreibe dein gewünschtes Bild:**") with gr.Row(): txt_input = gr.Textbox( placeholder="z.B. ultra realistic mountain landscape at sunrise...", lines=2, label="Prompt (Englisch)" ) with gr.Row(): with gr.Column(): txt_steps = gr.Slider( minimum=10, maximum=100, value=35, step=1, label="Inferenz-Schritte" ) with gr.Column(): txt_guidance = gr.Slider( minimum=1.0, maximum=20.0, value=7.5, step=0.5, label="Prompt-Stärke" ) generate_btn = gr.Button("Bild generieren", variant="primary") txt_output = gr.Image( label="Generiertes Bild", show_download_button=True, type="pil" ) # TEXT INTEGRATION FÜR TEXT-zu-BILD text_input_t2i, text_x_t2i, text_y_t2i, text_btn_t2i = create_text_integration_section_t2i() # VORSCHAU FÜR TEXT POSITION preview_t2i = gr.Image( label="Vorschau für Textposition (Klicken/Tippen um Position zu wählen)", interactive=True, show_download_button=False, type="pil" ) # CLICK HANDLER FÜR TEXT-zu-BILD preview_t2i.select( fn=capture_click, outputs=[text_x_t2i, text_y_t2i] ) # LIVE-TEXT-VORSCHAU FÜR TEXT-ZU-BILD text_input_t2i.change( fn=update_text_preview_t2i, inputs=[preview_t2i, text_input_t2i, text_x_t2i, text_y_t2i], outputs=preview_t2i ) text_x_t2i.change( fn=update_text_preview_t2i, inputs=[preview_t2i, text_input_t2i, text_x_t2i, text_y_t2i], outputs=preview_t2i ) text_y_t2i.change( fn=update_text_preview_t2i, inputs=[preview_t2i, text_input_t2i, text_x_t2i, text_y_t2i], outputs=preview_t2i ) # EVENT-HANDLER TEXT-zu-BILD generate_btn.click( fn=text_to_image, inputs=[txt_input, txt_steps, txt_guidance], outputs=[txt_output, preview_t2i], concurrency_limit=1 ) text_btn_t2i.click( fn=handle_text_integration_t2i, inputs=[txt_output, text_input_t2i, text_x_t2i, text_y_t2i], outputs=txt_output ) # === TAB: BILD ZU BILD === with gr.Tab("Bild zu Bild"): gr.Markdown("**Lade ein Bild hoch und beschreibe die gewünschte Veränderung:**") with gr.Row(): with gr.Column(): img_input = gr.Image( type="pil", label="Eingabebild", height=300, sources=["upload"] ) with gr.Column(): preview_output = gr.Image( label="Live-Vorschau mit Maske (Klicken/Tippen für Textposition)", height=300, interactive=True, show_download_button=False ) with gr.Row(): face_preserve = gr.Checkbox( label="Schutz", value=True, info="🟢 AN: Umgebung verändern | 🔴 AUS: Objekt verändern" ) with gr.Row(): with gr.Column(): bbox_x1 = gr.Slider(label="Links (x1)", minimum=0, maximum=512, value=100, step=1) with gr.Column(): bbox_y1 = gr.Slider(label="Oben (y1)", minimum=0, maximum=512, value=100, step=1) with gr.Row(): with gr.Column(): bbox_x2 = gr.Slider(label="Rechts (x2)", minimum=0, maximum=512, value=300, step=1) with gr.Column(): bbox_y2 = gr.Slider(label="Unten (y2)", minimum=0, maximum=512, value=300, step=1) with gr.Row(): with gr.Column(): img_prompt = gr.Textbox( placeholder="change background to beach with palm trees...", lines=2, label="Transformations-Prompt" ) with gr.Column(): img_neg_prompt = gr.Textbox( placeholder="blurry, deformed, ugly...", lines=2, label="Negativ-Prompt" ) with gr.Row(): with gr.Column(): strength_slider = gr.Slider(minimum=0.1, maximum=0.9, value=0.4, step=0.05, label="Veränderungs-Stärke") with gr.Column(): img_steps = gr.Slider(minimum=10, maximum=100, value=35, step=1, label="Inferenz-Schritte") with gr.Column(): img_guidance = gr.Slider(minimum=1.0, maximum=20.0, value=7.5, step=0.5, label="Prompt-Stärke") transform_btn = gr.Button("Bild transformieren", variant="primary") with gr.Row(): img_output = gr.Image( label="Transformiertes Bild", show_download_button=True, type="pil" ) # TEXT INTEGRATION FÜR BILD-zu-BILD text_input_i2i, text_x_i2i, text_y_i2i, target_selector, text_btn_i2i = create_text_integration_section_i2i() # CLICK HANDLER FÜR BILD-zu-BILD preview_output.select( fn=capture_click, outputs=[text_x_i2i, text_y_i2i] ) # LIVE-TEXT-VORSCHAU FÜR BILD-ZU-BILD text_input_i2i.change( fn=update_text_preview_i2i, inputs=[img_input, img_output, text_input_i2i, text_x_i2i, text_y_i2i, target_selector], outputs=preview_output ) text_x_i2i.change( fn=update_text_preview_i2i, inputs=[img_input, img_output, text_input_i2i, text_x_i2i, text_y_i2i, target_selector], outputs=preview_output ) text_y_i2i.change( fn=update_text_preview_i2i, inputs=[img_input, img_output, text_input_i2i, text_x_i2i, text_y_i2i, target_selector], outputs=preview_output ) target_selector.change( fn=update_text_preview_i2i, inputs=[img_input, img_output, text_input_i2i, text_x_i2i, text_y_i2i, target_selector], outputs=preview_output ) # EVENT-HANDLER BILD-zu-BILD img_input.change( fn=process_image_upload, inputs=[img_input], outputs=[preview_output, bbox_x1, bbox_y1, bbox_x2, bbox_y2] ) coordinate_inputs = [img_input, bbox_x1, bbox_y1, bbox_x2, bbox_y2, face_preserve] for coord in [bbox_x1, bbox_y1, bbox_x2, bbox_y2]: coord.change( fn=update_live_preview, inputs=coordinate_inputs, outputs=preview_output ) face_preserve.change( fn=update_live_preview, inputs=coordinate_inputs, outputs=preview_output ) transform_btn.click( fn=img_to_image, inputs=[ img_input, img_prompt, img_neg_prompt, strength_slider, img_steps, img_guidance, face_preserve, bbox_x1, bbox_y1, bbox_x2, bbox_y2 ], outputs=img_output, concurrency_limit=1 ) text_btn_i2i.click( fn=handle_text_integration_i2i, inputs=[img_input, img_output, text_input_i2i, text_x_i2i, text_y_i2i, target_selector], outputs=[img_input, img_output] ) # === START-BUTTON HANDLER === info_components = [child for child in demo.children if child != content_area] start_btn.click( fn=lambda: gr.update(visible=True), inputs=None, outputs=content_area ).then( fn=lambda: [gr.update(visible=False) for _ in info_components], inputs=None, outputs=info_components ) return demo if __name__ == "__main__": demo = main_ui() demo.queue() demo.launch( server_name="0.0.0.0", server_port=7860, max_file_size="10MB", show_error=True, share=False )