Update app.py

app.py

@@ -120,10 +120,6 @@ def estimate_depth(pil_image: Image.Image) ->Image.Image:
 def generate_image_for_gradio(
     prompt: str,
     input_image_for_depth: Image.Image,
-    num_inference_steps: int = 25,
-    guidance_scale: float = 8.0,
-    seed: int = None,
-    resolution: int = 512
 ) -> Image.Image:
 
     global pipeline
@@ -140,24 +136,34 @@ def generate_image_for_gradio(
 
     print(f"Generating image for prompt: '{prompt}'")
 
-
+    negative_prompt = "lowres, watermark, banner, logo, watermark, contactinfo, text, deformed, blurry, blur, out of focus, out of frame, surreal, ugly"
     control_image = depth_map_pil.convert("RGB")
-    control_image = control_image.resize((resolution, resolution), Image.LANCZOS)
+    control_image = control_image.resize((512, 512), Image.LANCZOS)
 
     input_image_for_pipeline = [control_image]
 
     generator = None
-    if seed is None:
-        seed = random.randint(0, 100000)
+    # if seed is None:
+    seed = random.randint(0, 100000)
     generator = torch.Generator(device=DEVICE).manual_seed(seed)
 
 
+    # with torch.no_grad():
+    #     generated_images = pipeline(
+    #         prompt,
+    #         image=input_image_for_pipeline,
+    #         num_inference_steps=num_inference_steps,
+    #         guidance_scale=guidance_scale,
+    #         generator=generator,
+    #     ).images
+
     with torch.no_grad():
         generated_images = pipeline(
             prompt,
+            negative_prompt=negative_prompt, # Pass negative prompt
             image=input_image_for_pipeline,
-            num_inference_steps=num_inference_steps,
-            guidance_scale=guidance_scale,
+            num_inference_steps=50,
+            guidance_scale=0.85,
             generator=generator,
         ).images
 
@@ -166,15 +172,28 @@ def generate_image_for_gradio(
 
 
 
+# iface = gr.Interface(
+#     fn=generate_image_for_gradio,
+#     inputs=[
+#         gr.Textbox(label="Prompt", value="a high-quality photo of a modern interior design"),
+#         gr.Image(type="pil", label="Input Image (for Depth Estimation)"),
+#         gr.Slider(minimum=10, maximum=100, value=25, step=1, label="Inference Steps"),
+#         gr.Slider(minimum=1.0, maximum=20.0, value=8.0, step=0.5, label="Guidance Scale"),
+#         gr.Number(label="Seed (optional, leave blank for random)", value=None),
+#         gr.Number(label="Resolution", value=512, interactive=False)
+#     ],
+#     outputs=gr.Image(type="pil", label="Generated Image"),
+#     title="Stable Diffusion ControlNet Depth Demo (with Depth Estimation)",
+#     description="Upload an input image, and the app will estimate its depth map, then use it with your prompt to generate a new image. This allows for structural guidance from your input photo.",
+#     allow_flagging="never",
+#     live=False,
+#     theme=Soft(),
+
 iface = gr.Interface(
     fn=generate_image_for_gradio,
     inputs=[
-        gr.Textbox(label="Prompt", value="a high-quality photo of a modern interior design"),
         gr.Image(type="pil", label="Input Image (for Depth Estimation)"),
-        gr.Slider(minimum=10, maximum=100, value=25, step=1, label="Inference Steps"),
-        gr.Slider(minimum=1.0, maximum=20.0, value=8.0, step=0.5, label="Guidance Scale"),
-        gr.Number(label="Seed (optional, leave blank for random)", value=None),
-        gr.Number(label="Resolution", value=512, interactive=False)
+        gr.Textbox(label="Prompt", value="a high-quality photo of a modern interior design"),
     ],
     outputs=gr.Image(type="pil", label="Generated Image"),
     title="Stable Diffusion ControlNet Depth Demo (with Depth Estimation)",