Fix dtype issue for legacy pipeline

app.py

@@ -80,6 +80,7 @@ finally:
         device = "cpu"
 
 import contextlib
+
 autocast = contextlib.nullcontext
 
 with open("config.yaml", "r") as yaml_in:
@@ -290,7 +291,7 @@ class StableDiffusionInpaint:
                     revision="fp16",
                     torch_dtype=torch.float16,
                     use_auth_token=token,
-                    vae=vae
+                    vae=vae,
                 )
             else:
                 inpaint = StableDiffusionInpaintPipeline.from_pretrained(
@@ -458,7 +459,7 @@ class StableDiffusion:
                     revision="fp16",
                     torch_dtype=torch.float16,
                     use_auth_token=token,
-                    vae = vae
+                    vae=vae,
                 )
             else:
                 text2img = StableDiffusionPipeline.from_pretrained(
@@ -482,11 +483,13 @@ class StableDiffusion:
                     revision="fp16",
                     torch_dtype=torch.float16,
                     use_auth_token=token,
-                    vae=vae
+                    vae=vae,
                 ).to(device)
             else:
                 inpaint = StableDiffusionInpaintPipeline.from_pretrained(
-                    "runwayml/stable-diffusion-inpainting", use_auth_token=token, vae=vae
+                    "runwayml/stable-diffusion-inpainting",
+                    use_auth_token=token,
+                    vae=vae,
                 ).to(device)
             text2img_unet.to(device)
             text2img = StableDiffusionPipeline(
@@ -657,10 +660,11 @@ class StableDiffusion:
             init_image = Image.fromarray(img)
             mask_image = Image.fromarray(mask)
             # mask_image=mask_image.filter(ImageFilter.GaussianBlur(radius = 8))
-            if True:
-                input_image = init_image.resize(
-                    (process_width, process_height), resample=SAMPLING_MODE
-                )
+            input_image = init_image.resize(
+                (process_width, process_height), resample=SAMPLING_MODE
+            )
+            if self.inpainting_model:
+
                 images = inpaint_func(
                     prompt=prompt,
                     init_image=input_image,
@@ -670,6 +674,17 @@ class StableDiffusion:
                     mask_image=mask_image.resize((process_width, process_height)),
                     **extra_kwargs,
                 )["images"]
+            else:
+                with autocast("cuda"):
+                    images = inpaint_func(
+                        prompt=prompt,
+                        init_image=input_image,
+                        image=input_image,
+                        width=process_width,
+                        height=process_height,
+                        mask_image=mask_image.resize((process_width, process_height)),
+                        **extra_kwargs,
+                    )["images"]
         else:
             if True:
                 images = text2img(
@@ -740,20 +755,20 @@ def run_outpaint(
     pil = Image.open(io.BytesIO(data))
     if interrogate_mode:
         if "interrogator" not in model:
-            model["interrogator"]=Interrogator()
+            model["interrogator"] = Interrogator()
         interrogator = model["interrogator"]
-        img=np.array(pil)[:,:,0:3]
-        mask=np.array(pil)[:,:,-1]
+        img = np.array(pil)[:, :, 0:3]
+        mask = np.array(pil)[:, :, -1]
         x, y = np.nonzero(mask)
-        if len(x)>0:
+        if len(x) > 0:
             x0, x1 = x.min(), x.max() + 1
             y0, y1 = y.min(), y.max() + 1
-            img=img[x0:x1,y0:y1,:]
-        pil=Image.fromarray(img)
+            img = img[x0:x1, y0:y1, :]
+        pil = Image.fromarray(img)
         interrogate_ret = interrogator.interrogate(pil)
         return (
             gr.update(value=",".join([sel_buffer_str]),),
-            gr.update(label="Prompt",value=interrogate_ret),
+            gr.update(label="Prompt", value=interrogate_ret),
             state,
         )
     width, height = pil.size
@@ -995,7 +1010,10 @@ with blocks as demo:
             except Exception as e:
                 print(e)
                 return {token: gr.update(value=str(e))}
-            if model_choice in [ModelChoice.INPAINTING.value,ModelChoice.INPAINTING_IMG2IMG.value]:
+            if model_choice in [
+                ModelChoice.INPAINTING.value,
+                ModelChoice.INPAINTING_IMG2IMG.value,
+            ]:
                 init_val = "cv2_ns"
             else:
                 init_val = "patchmatch"
@@ -1009,7 +1027,7 @@ with blocks as demo:
                 upload_button: gr.update(value="Upload Image"),
                 model_selection: gr.update(visible=False),
                 model_path_input: gr.update(visible=False),
-                init_mode: gr.update(value=init_val)
+                init_mode: gr.update(value=init_val),
             }
 
         setup_button.click(
@@ -1032,7 +1050,7 @@ with blocks as demo:
                 upload_button,
                 model_selection,
                 model_path_input,
-                init_mode
+                init_mode,
             ],
             _js=setup_button_js,
         )

utils.py

@@ -93,8 +93,8 @@ def edge_pad(img, mask, mode=1):
 
 
 def perlin_noise(img, mask):
-    lin_x = np.linspace(0, 5, mask.shape[0], endpoint=False)
-    lin_y = np.linspace(0, 5, mask.shape[1], endpoint=False)
+    lin_x = np.linspace(0, 5, mask.shape[1], endpoint=False)
+    lin_y = np.linspace(0, 5, mask.shape[0], endpoint=False)
     x, y = np.meshgrid(lin_x, lin_y)
     avg = img.mean(axis=0).mean(axis=0)
     # noise=[((perlin(x, y)+1)*128+avg[i]).astype(np.uint8) for i in range(3)]