import torch from diffusers import StableDiffusionControlNetPipeline, ControlNetModel from controlnet_aux import OpenposeDetector from PIL import Image import random import cv2 import numpy as np import gradio as gr class ControlNetProgressCallback: def __init__(self, progress, total_steps): self.progress = progress self.total_steps = total_steps self.current_step = 0 def __call__(self, pipe, step_index, timestep, callback_kwargs): self.current_step = step_index + 1 progress_percentage = self.current_step / self.total_steps # Fortschritt aktualisieren if self.progress is not None: self.progress(progress_percentage, desc=f"ControlNet: Schritt {self.current_step}/{self.total_steps}") print(f"ControlNet Fortschritt: {self.current_step}/{self.total_steps} ({progress_percentage:.1%})") return callback_kwargs class ControlNetProcessor: def __init__(self, device="cuda", torch_dtype=torch.float32): self.device = device self.torch_dtype = torch_dtype self.pose_detector = None self.controlnet_openpose = None self.controlnet_canny = None self.pipe_openpose = None self.pipe_canny = None def load_pose_detector(self): """Lädt nur den Pose-Detector""" if self.pose_detector is None: print("Loading Pose Detector...") try: self.pose_detector = OpenposeDetector.from_pretrained("lllyasviel/ControlNet") except Exception as e: print(f"Warnung: Pose-Detector konnte nicht geladen werden: {e}") return self.pose_detector def extract_pose_simple(self, image): """Einfache Pose-Extraktion ohne komplexe Abhängigkeiten""" try: img_array = np.array(image.convert("RGB")) edges = cv2.Canny(img_array, 100, 200) pose_image = Image.fromarray(edges).convert("RGB") print("⚠️ Verwende Kanten-basierte Pose-Approximation") return pose_image except Exception as e: print(f"Fehler bei einfacher Pose-Extraktion: {e}") return image.convert("RGB").resize((512, 512)) def extract_pose(self, image): """Extrahiert Pose-Map aus Bild mit Fallback""" try: detector = self.load_pose_detector() if detector is None: return self.extract_pose_simple(image) pose_image = detector(image, hand_and_face=True) return pose_image except Exception as e: print(f"Fehler bei Pose-Extraktion: {e}") return self.extract_pose_simple(image) def extract_canny_edges(self, image): """Extrahiert Canny Edges für Umgebungserhaltung""" try: img_array = np.array(image.convert("RGB")) # Canny Edge Detection gray = cv2.cvtColor(img_array, cv2.COLOR_RGB2GRAY) edges = cv2.Canny(gray, 100, 200) # Zu 3-Kanal Bild konvertieren edges_rgb = cv2.cvtColor(edges, cv2.COLOR_GRAY2RGB) edges_image = Image.fromarray(edges_rgb) print("✅ Canny Edge für Umgebungserhaltung erstellt") return edges_image except Exception as e: print(f"Fehler bei Canny Edge Extraction: {e}") return image.convert("RGB").resize((512, 512)) def load_controlnet_pipeline(self, controlnet_type="openpose"): """Lädt die passende ControlNet Pipeline""" if controlnet_type == "openpose": if self.pipe_openpose is None: print("Loading OpenPose ControlNet pipeline...") try: self.controlnet_openpose = ControlNetModel.from_pretrained( "lllyasviel/sd-controlnet-openpose", torch_dtype=self.torch_dtype ) self.pipe_openpose = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", controlnet=self.controlnet_openpose, torch_dtype=self.torch_dtype, safety_checker=None, requires_safety_checker=False ).to(self.device) from diffusers import EulerAncestralDiscreteScheduler self.pipe_openpose.scheduler = EulerAncestralDiscreteScheduler.from_config(self.pipe_openpose.scheduler.config) self.pipe_openpose.enable_attention_slicing() print("✅ OpenPose ControlNet pipeline loaded successfully!") except Exception as e: print(f"Fehler beim Laden von OpenPose ControlNet: {e}") raise return self.pipe_openpose elif controlnet_type == "canny": if self.pipe_canny is None: print("Loading Canny ControlNet pipeline...") try: self.controlnet_canny = ControlNetModel.from_pretrained( "lllyasviel/sd-controlnet-canny", torch_dtype=self.torch_dtype ) self.pipe_canny = StableDiffusionControlNetPipeline.from_pretrained( "runwayml/stable-diffusion-v1-5", controlnet=self.controlnet_canny, torch_dtype=self.torch_dtype, safety_checker=None, requires_safety_checker=False ).to(self.device) from diffusers import EulerAncestralDiscreteScheduler self.pipe_canny.scheduler = EulerAncestralDiscreteScheduler.from_config(self.pipe_canny.scheduler.config) self.pipe_canny.enable_attention_slicing() print("✅ Canny ControlNet pipeline loaded successfully!") except Exception as e: print(f"Fehler beim Laden von Canny ControlNet: {e}") raise return self.pipe_canny def generate_with_controlnet( self, image, prompt, negative_prompt, steps, guidance_scale, controlnet_strength, progress=None, keep_environment=False ): """Generiert Bild mit ControlNet und Fortschrittsanzeige""" try: # --- KORREKTE LOGIK --- if keep_environment: # UMGEBUNG BEIBEHALTEN, PERSON ÄNDERN controlnet_type = "canny" # ✅ Canny behält Umgebung print("🎯 ControlNet Modus: Umgebung beibehalten (Canny Edge)") conditioning_image = self.extract_canny_edges(image) inpaint_input = image # ✅ ORIGINALBILD für Inpaint! else: # PERSON BEIBEHALTEN, UMGEBUNG ÄNDERN controlnet_type = "openpose" # ✅ OpenPose behält Person print("🎯 ControlNet Modus: Person beibehalten (OpenPose)") conditioning_image = self.extract_pose(image) inpaint_input = conditioning_image # ✅ POSE-MAP für Inpaint pipe = self.load_controlnet_pipeline(controlnet_type) # Zufälliger Seed seed = random.randint(0, 2**32 - 1) generator = torch.Generator(device=self.device).manual_seed(seed) print(f"ControlNet Seed: {seed}") # Fortschritt-Callback callback = ControlNetProgressCallback(progress, int(steps)) if progress is not None else None print("🔄 ControlNet: Starte Pipeline...") # ControlNet Generierung result = pipe( prompt=prompt, image=conditioning_image, negative_prompt=negative_prompt, num_inference_steps=int(steps), guidance_scale=guidance_scale, generator=generator, controlnet_conditioning_scale=controlnet_strength, height=512, width=512, output_type="pil", callback_on_step_end=callback, callback_on_step_end_tensor_inputs=[], ) # Debug-Ausgabe Scheduler Steps try: scheduler = pipe.scheduler if hasattr(scheduler, "timesteps"): actual_steps = len(scheduler.timesteps) print(f"🎯 CONTROLNET TATSÄCHLICHE STEPS: {actual_steps} (von {steps} angefordert)") except Exception as e: print(f"⚠️ Konnte ControlNet Scheduler-Info nicht auslesen: {e}") print("✅ ControlNet abgeschlossen!") # DREI Werte zurückgeben return result.images[0], conditioning_image, inpaint_input except Exception as e: print(f"❌ Fehler in ControlNet: {e}") import traceback traceback.print_exc() error_image = image.convert("RGB").resize((512, 512)) return error_image, error_image, error_image def prepare_inpaint_input(self, image, keep_environment=False): """ Bereitet das Input-Bild für Inpaint vor Rückgabe: (image_für_inpaint, conditioning_info) """ if keep_environment: # PERSON ÄNDERN: Originalbild an Inpaint übergeben print("🎯 Inpaint: Übergebe Originalbild (Person ändern)") return image, {"type": "original", "image": image} else: # UMGEBUNG ÄNDERN: Pose-Map an Inpaint übergeben print("🎯 Inpaint: Übergebe Pose-Map (Umgebung ändern)") pose_image = self.extract_pose(image) return pose_image, {"type": "pose", "image": pose_image} # Globale Instanz device = "cuda" if torch.cuda.is_available() else "cpu" torch_dtype = torch.float16 if device == "cuda" else torch.float32 controlnet_processor = ControlNetProcessor(device=device, torch_dtype=torch_dtype)