Make ControlNet work

app.py

@@ -3,21 +3,19 @@ import gradio as gr
 import numpy as np
 import torch
 
-from diffusers import DiffusionPipeline, StableDiffusionControlNetPipeline
+from diffusers import ControlNetModel, DiffusionPipeline, StableDiffusionControlNetPipeline
 from PIL import Image
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
 low_threshold = 100
 high_threshold = 200
 
 def generate(
     prompt, negative_prompt, num_inference_steps, width, height, guidance_scale, seed, input_image
 ):
+    generator = torch.manual_seed(seed)
+
     if input_image is None:
-        pipeline = DiffusionPipeline.from_pretrained("Lykon/DreamShaper")
-        pipeline = pipeline.to(device)
-        generator = torch.Generator(device=device).manual_seed(seed)
+        pipeline = DiffusionPipeline.from_pretrained("Lykon/DreamShaper", torch_dtype=torch.float16)
 
         return pipeline(
             prompt=prompt,
@@ -34,7 +32,19 @@ def generate(
     image = np.concatenate([image, image, image], axis=2)
     canny_image = Image.fromarray(image)
 
-    return canny_image
+    controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
+    pipeline = StableDiffusionControlNetPipeline.from_pretrained("Lykon/DreamShaper", controlnet=controlnet, torch_dtype=torch.float16)
+
+    return pipeline(
+        prompt=prompt,
+        negative_prompt=negative_prompt,
+        num_inference_steps=num_inference_steps,
+        width=width,
+        height=height,
+        guidance_scale=guidance_scale,
+        generator=generator,
+        image=canny_image,
+    ).images[0]
 
 iface = gr.Interface(
     fn=generate,