Update app.py

app.py

@@ -1,6 +1,8 @@
 import torch
 import gradio as gr
 import spaces
+import random
+import numpy as np
 
 from pipeline import ChatsSDXLPipeline
 from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
@@ -11,6 +13,7 @@ from PIL import Image
 logging.set_verbosity_error()
 
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+MAX_SEED = np.iinfo(np.int32).max
 
 feature_extractor = CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32")
 safety_checker = StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker")
@@ -20,67 +23,101 @@ pipe = ChatsSDXLPipeline.from_pretrained(
         "AIDC-AI/CHATS",
         safety_checker=safety_checker,
         feature_extractor=feature_extractor,
-        torch_dtype=torch.float16
+        torch_dtype=torch.bfloat16
 )
 pipe.to(DEVICE)
 
-@spaces.GPU
-def generate(prompt, steps=50, guidance_scale=7.5, height=768, width=512):
-  output = pipe(
-      prompt=prompt,
-      num_inference_steps=steps,
-      guidance_scale=guidance_scale,
-      height=height,
-      width=width,
-      seed=0
-  )
-  return output['images']
-  # image = output['images'][0]
-  # image = Image.fromarray(image)
-  # return image
+@spaces.GPU(duration=75)
+def generate(prompt, seed, randomize_seed=False, steps=50, guidance_scale=7.5):
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)
 
-with gr.Blocks(title="🔥 CHATS-SDXL Demo") as demo:
-    gr.Markdown(
-        "## CHATS-SDXL Text-to-Image Demo\n\n"
-        "Enter your prompt and click **Generate Image**. All NSFW content will be automatically filtered."
+    output = pipe(
+        prompt=prompt,
+        num_inference_steps=steps,
+        guidance_scale=guidance_scale,
+        seed=seed
     )
-    with gr.Row():
-        prompt_input = gr.Textbox(
-            label="Prompt",
-            placeholder="Enter your description here...",
-            lines=2,
-        )
-    with gr.Row():
-        steps_slider = gr.Slider(
-            minimum=1, maximum=100, value=50, step=1,
-            label="Inference Steps"
-        )
-        scale_slider = gr.Slider(
-            minimum=1.0, maximum=14.0, value=5.0, step=0.1,
-            label="Guidance Scale"
-        )
-    with gr.Row():
-        height_slider = gr.Slider(
-            minimum=64, maximum=2048, value=1024, step=64, 
-            label="Image Height"
-        )
-        width_slider = gr.Slider(
-            minimum=64, maximum=2048, value=1024, step=64, 
-            label="Image Width"
+    return output['images']
+
+examples = [
+    "The image is a digital art headshot of an owlfolk character with high detail and dramatic lighting",
+    "Solar punk vehicle in a bustling city",
+    "An elderly woman poses for a high fashion photoshoot in colorful, patterned clothes with a cyberpunk 2077 vibe",
+]
+
+css="""
+#col-container {
+    margin: 0 auto;
+    max-width: 520px;
+}
+"""
+
+with gr.Blocks(css=css) as demo:
+    
+    with gr.Column(elem_id="col-container"):
+        gr.Markdown(f"""# CHATS-SDXL
+SDXL diffusion models finetuned using preference optimization framework CHATS. [[paper] (https://arxiv.org/pdf/2502.12579)] [[code](https://github.com/AIDC-AI/CHATS)] [[model](https://huggingface.co/AIDC-AI/CHATS)]
+        """)
+        
+        with gr.Row():
+            
+            prompt = gr.Text(
+                label="Prompt",
+                show_label=False,
+                max_lines=1,
+                placeholder="Enter your prompt here",
+                container=False,
+            )
+            
+            run_button = gr.Button("Run", scale=0)
+        
+        result = gr.Image(label="Result", show_label=False)
+        
+        with gr.Accordion("Advanced Settings", open=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():
+
+                guidance_scale = gr.Slider(
+                    label="Guidance Scale",
+                    minimum=1,
+                    maximum=14,
+                    step=0.1,
+                    value=5.0,
+                )
+  
+                num_inference_steps = gr.Slider(
+                    label="Number of inference steps",
+                    minimum=1,
+                    maximum=100,
+                    step=1,
+                    value=50,
+                )
+        
+        gr.Examples(
+            examples = examples,
+            fn = generate,
+            inputs = [prompt],
+            outputs = [result],
+            cache_examples="lazy"
         )
-    generate_button = gr.Button("Generate Image")
-    gallery = gr.Gallery(
-        label="Generated Images",
-        show_label=False,
-        columns=2,     
-        elem_id="gallery"
-    )
 
-    generate_button.click(
-        fn=generate,
-        inputs=[prompt_input, steps_slider, scale_slider, height_slider, width_slider],
-        outputs=[gallery],
+    gr.on(
+        triggers=[run_button.click, prompt.submit],
+        fn = generate,
+        inputs = [prompt, seed, randomize_seed, num_inference_steps, guidance_scale],
+        outputs = [result]
     )
 
-if __name__ == "__main__":
-    demo.launch()
+if __name__ == '__main__':
+    demo.launch()