Better

app.py

@@ -2,9 +2,10 @@ import gradio as gr
 import torch
 
 from diffusers import AutoPipelineForInpainting, UNet2DConditionModel
+import diffusers
 from share_btn import community_icon_html, loading_icon_html, share_js
 
-unet = UNet2DConditionModel.from_pretrained("valhalla/sdxl-inpaint-ema")
+unet = UNet2DConditionModel.from_pretrained("valhalla/sdxl-inpaint-ema", torch_dtype=torch.float16)
 pipe = AutoPipelineForInpainting.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", unet=unet, torch_dtype=torch.float16, variant="fp16").to("cuda")
 
 def read_content(file_path: str) -> str:
@@ -15,12 +16,23 @@ def read_content(file_path: str) -> str:
 
     return content
 
-def predict(dict, prompt="", guidance_scale=7.5, steps=20, strength=1.0):
+def predict(dict, prompt="", guidance_scale=7.5, steps=20, strength=1.0, scheduler="EulerDiscreteScheduler"):
+
+    scheduler_class_name = scheduler.split("")[0]
+    scheduler = getattr(diffusers, scheduler_class_name)
+
+    add_kwargs = {}
+    if len(scheduler.split("")) > 1:
+        add_kwargs["use_karras"] = True
+    if len(scheduler.split("")) > 2:
+        add_kwargs["algorithm_type"] = "sde-dpmsolver++"
+
+    pipe.scheduler = scheduler.from_config(pipe.scheduler.config, **add_kwargs)
     
     init_image = dict["image"].convert("RGB").resize((1024, 1024))
     mask = dict["mask"].convert("RGB").resize((1024, 1024))
     
-    output = pipe(prompt = prompt, image=init_image, mask_image=mask, guidance_scale=guidance_scale, steps=int(steps), strength=strength)
+    output = pipe(prompt = prompt, image=init_image, mask_image=mask, guidance_scale=guidance_scale, num_inference_steps=int(steps), strength=strength)
     
     return output.images[0], gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
 
@@ -79,6 +91,8 @@ with image_blocks as demo:
                         guidance_scale = gr.Number(value=7.5, minimum=1.0, maximum=20.0, step=0.1, label="guidance_scale")
                         steps = gr.Number(value=20, minimum=10, maximum=50, step=0.1, label="steps")
                         strength = gr.Number(value=1.0, minimum=0.0, maximum=1.0, step=0.05, label="strength")
+                        schedulers = ["DEISMultistepScheduler", "HeunDiscreteScheduler", "EulerDiscreteScheduler", "DPMSolverMultistepScheduler", "DPMSolverMultistepScheduler Karras", "DPMSolverMultistepScheduler Karras SDE"]
+                        scheduler = gr.Dropdown(choices=schedulers, value="EulerDiscreteScheduler")
                         
                         btn = gr.Button("Inpaint!").style(
                             margin=False,
@@ -93,13 +107,13 @@ with image_blocks as demo:
                         share_button = gr.Button("Share to community", elem_id="share-btn", visible=False)
             
 
-            btn.click(fn=predict, inputs=[image, prompt, guidance_scale, steps, strength], outputs=[image_out, community_icon, loading_icon, share_button])
+            btn.click(fn=predict, inputs=[image, prompt, guidance_scale, steps, strength, scheduler], outputs=[image_out, community_icon, loading_icon, share_button])
             share_button.click(None, [], [], _js=share_js)
 
             gr.HTML(
                 """
                     <div class="footer">
-                        <p>Model by <a href="https://huggingface.co/diffusers" style="text-decoration: underline;" target="_blank">RunwayML</a> - Gradio Demo by 🤗 Hugging Face
+                        <p>Model by <a href="https://huggingface.co/diffusers" style="text-decoration: underline;" target="_blank">Diffusers</a> - Gradio Demo by 🤗 Hugging Face
                         </p>
                     </div>
                 """