Changes to be committed:

app.py

@@ -1,146 +1,112 @@
+import spaces
 import gradio as gr
-import numpy as np
-import random
-from diffusers import DiffusionPipeline
 import torch
+from PIL import Image
+from diffusers import DiffusionPipeline, StableDiffusionXLImg2ImgPipeline, AutoencoderKL
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
+device = "cuda"
 
-if torch.cuda.is_available():
-    torch.cuda.max_memory_allocated(device=device)
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16", use_safetensors=True)
-    pipe.enable_xformers_memory_efficient_attention()
-    pipe = pipe.to(device)
-else: 
-    pipe = DiffusionPipeline.from_pretrained("stabilityai/sdxl-turbo", use_safetensors=True)
-    pipe = pipe.to(device)
+base_model_id = "stabilityai/stable-diffusion-xl-base-1.0"
+vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+refiner_id = "stabilityai/stable-diffusion-xl-refiner-1.0"
 
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+base_pipeline = DiffusionPipeline.from_pretrained(
+    base_model_id, 
+    torch_dtype = torch.float16, 
+    variant = "fp16", 
+    use_safetensors = True
+).to(device)
 
-def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
+refiner = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+    refiner_id, 
+    text_encoder_2 = base_pipeline.text_encoder_2, 
+    vae = vae,  
+    torch_dtype = torch.float16, 
+    variant = "fp16", 
+    use_safetensors = True
+).to(device)
 
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-        
-    generator = torch.Generator().manual_seed(seed)
+
+
+SAMPLER_MAP = {
+    "DPM++ Karras SDE": lambda config: DPMSolverMultistepScheduler.from_config(config, use_karras=True, algorithm_type="sde-dpmsolver++"),
+    "Euler": lambda config: EulerDiscreteScheduler.from_config(config),
+}
+
+
+
+@spaces.GPU(duration=120)
+def generate(
+    prompt, 
+    negative_prompt, 
+    num_inference_steps, 
+    denoising_switch, 
+    width, height, 
+    guidance_scale
+):
     
-    image = pipe(
+    base_processed_image = base_pipeline(
         prompt = prompt, 
-        negative_prompt = negative_prompt,
+        negative_prompt = negative_prompt, 
+        num_inference_steps = num_inference_steps, 
+        denoising_end = denoising_switch, 
+        width = width, 
+        height = height, 
         guidance_scale = guidance_scale, 
+        output_type = "latent"
+    ).images
+
+    generated_image = refiner(
+        prompt = prompt, 
+        negative_prompt = negative_prompt, 
         num_inference_steps = num_inference_steps, 
+        denoising_start = denoising_switch, 
         width = width, 
-        height = height,
-        generator = generator
-    ).images[0] 
-    
-    return image
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css="""
-#col-container {
-    margin: 0 auto;
-    max-width: 520px;
-}
-"""
+        height = height, 
+        guidance_scale = guidance_scale, 
+        image = base_processed_image
+    ).images[0]
 
-if torch.cuda.is_available():
-    power_device = "GPU"
-else:
-    power_device = "CPU"
+    return generated_image
 
-with gr.Blocks(css=css) as demo:
-    
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(f"""
-        # Text-to-Image Gradio Template
-        Currently running on {power_device}.
-        """)
-        
+
+def create_ui():
+    with gr.Blocks() as demo:
         with gr.Row():
-            
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-            
-            run_button = gr.Button("Run", scale=0)
-        
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-            
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-            
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-            
-            with gr.Row():
-                
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=512,
-                )
-                
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=512,
-                )
-            
-            with gr.Row():
-                
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,
-                )
-                
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=12,
-                    step=1,
-                    value=2,
-                )
-        
-        gr.Examples(
-            examples = examples,
-            inputs = [prompt]
+            base_model = gr.Radio(label="Base model", choices=[base_model_id], value=base_model_id, interactive=False)
+            refiner_model = gr.Radio(label="Refiner model", choices=[refiner_id], value=refiner_id, interactive=False)
+        with gr.Row():
+            prompt = gr.Textbox(lines=3)
+            negative_prompt = gr.Textbox(lines=3)
+        with gr.Row():
+            with gr.Column():
+                num_inference_steps = gr.Slider(label="Steps", minimum=1, maximum=50, step=1, value=30)
+                denoising_switch = gr.Slider(label="Denoising Switch", minimum=0.01, maximum=1, step=0.01, value=0.8)
+                width = gr.Slider(label="Width", minimum=64, maximum=2048, step=16, value=1024)
+                height = gr.Slider(label="Height", minimum=64, maximum=2048, step=16, value=1024)
+                guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.1, maximum=30, step=0.1, value=7.5)
+            with gr.Column():
+                output_image = gr.Image(interactive=False)
+                generate_button = gr.Button("Run", variant="primary")
+    
+        generate_button.click(
+            generate, 
+            inputs=[
+                prompt, 
+                negative_prompt, 
+                num_inference_steps, 
+                denoising_switch, 
+                width, height, 
+                guidance_scale
+            ], 
+            outputs=[output_image]
         )
 
-    run_button.click(
-        fn = infer,
-        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result]
-    )
+    return demo
+
 
-demo.queue().launch()
+if __name__ == "__main__":
+    gradio_app = create_ui()
+    gradio_app.launch(
+        share = True
+    )