import os print("Current working directory:", os.getcwd()) import gradio as gr import tempfile from SDXLImageGenerator import SDXLImageGenerator # Import your existing class import sys from Image3DProcessor import Image3DProcessor # Import your 3D processing class class ControlNetProcessor: def controlnet_image(self, image): # Placeholder for ControlNet processing (e.g., returning a processed image or placeholder text) return image # Returning the image for further processing class VideoGenerator: def __init__(self, model_cfg_path, model_repo_id, model_filename): # Initialize the Image3DProcessor self.processor = Image3DProcessor(model_cfg_path, model_repo_id, model_filename) def generate_3d_video(self, image): # Preprocess the image first processed_image = self.processor.preprocess(image) # Then pass it to reconstruct_and_export mesh_data, video_data = self.processor.reconstruct_and_export(processed_image) return mesh_data, video_data class GradioApp: def __init__(self): self.sdxl_generator = SDXLImageGenerator() # Use your existing class self.controlnet_processor = ControlNetProcessor() # Initialize VideoGenerator with required paths and details self.video_generator = VideoGenerator( model_cfg_path="/home/user/app/splatter-image/gradio_config.yaml", model_repo_id="szymanowiczs/splatter-image-multi-category-v1", model_filename="model_latest.pth" ) def full_pipeline(self, prompt): initial_image = self.sdxl_generator.generate_images([prompt])[0] # controlled_image = self.controlnet_processor.controlnet_image(initial_image) mesh_data, video_data = self.video_generator.generate_3d_video(initial_image) # Create temporary files to display mesh and video content with tempfile.NamedTemporaryFile(delete=False, suffix=".ply") as mesh_file: mesh_file.write(mesh_data) mesh_path = mesh_file.name with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as video_file: video_file.write(video_data) video_path = video_file.name return initial_image, mesh_path, video_path def launch(self): interface = gr.Interface( fn=self.full_pipeline, inputs=gr.Textbox(label="Input Prompt"), outputs=[ gr.Image(label="Generated Image"), gr.File(label="3D Mesh (.ply)"), gr.Video(label="3D Model Video") ], title="SDXL to ControlNet to 3D Pipeline", description="Generate an image using SDXL, refine it with ControlNet, and generate a 3D video output." ) interface.launch(share=True) # Added `share=True` for public link if __name__ == "__main__": app = GradioApp() app.launch()