Upload runtemporalnetxl.py

runtemporalnetxl.py

@@ -61,5 +61,55 @@ else:
     initial_frame_path = os.path.join(frames_dir, "frame0000.png")
     last_generated_image = load_image(initial_frame_path)
 
-# ... (rest of the script remains unchanged)
+base_model_path = "stabilityai/stable-diffusion-xl-base-1.0"
+controlnet1_path = "CiaraRowles/controlnet-temporalnet-sdxl-1.0"
+controlnet2_path = "diffusers/controlnet-canny-sdxl-1.0"
 
+controlnet = [
+    ControlNetModel.from_pretrained(controlnet1_path, torch_dtype=torch.float16),
+    ControlNetModel.from_pretrained(controlnet2_path, torch_dtype=torch.float16)
+]
+#controlnet = ControlNetModel.from_pretrained(controlnet2_path, torch_dtype=torch.float16)
+
+pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
+    base_model_path, controlnet=controlnet, torch_dtype=torch.float16
+)
+
+#pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+#pipe.enable_xformers_memory_efficient_attention()
+pipe.enable_model_cpu_offload()
+
+generator = torch.manual_seed(7)
+
+# Loop over the saved frames in numerical order
+frame_files = sorted(os.listdir(frames_dir), key=frame_number)
+
+for i, frame_file in enumerate(frame_files):
+    # Use the original video frame to create Canny edge-detected image as the conditioning image for the first ControlNetModel
+    control_image_path = os.path.join(frames_dir, frame_file)
+    control_image = load_image(control_image_path)
+    
+    canny_image = np.array(control_image)
+    canny_image = cv2.Canny(canny_image, 25, 200)
+    canny_image = canny_image[:, :, None]
+    canny_image = np.concatenate([canny_image, canny_image, canny_image], axis=2)
+    canny_image = Image.fromarray(canny_image)
+
+    # Generate image
+    image = pipe(
+       prompt, num_inference_steps=20, generator=generator, image=[last_generated_image, canny_image], controlnet_conditioning_scale=[0.6, 0.7]
+       #prompt, num_inference_steps=20, generator=generator, image=canny_image, controlnet_conditioning_scale=0.5
+    ).images[0]
+    
+    # Save the generated image to output folder
+    output_path = os.path.join(output_frames_dir, f"output{str(i).zfill(4)}.png")
+    image.save(output_path)
+
+    # Save the Canny image for reference
+    canny_image_path = os.path.join(output_frames_dir, f"outputcanny{str(i).zfill(4)}.png")
+    canny_image.save(canny_image_path)
+
+    # Update the last_generated_image with the newly generated image for the next iteration
+    last_generated_image = image
+
+    print(f"Saved generated image for frame {i} to {output_path}")