uuu

app.py

@@ -1,10 +1,19 @@
 import os
 import requests
+import torch
+import gradio as gr
+import spaces
+from PIL import Image
+from huggingface_hub import login
+from diffusers.utils import load_image
+
+from models.transformer_sd3 import SD3Transformer2DModel
+from pipeline_stable_diffusion_3_ipa import StableDiffusion3Pipeline
 
+# Download IP Adapter if not exists
 url = "https://huggingface.co/InstantX/SD3.5-Large-IP-Adapter/resolve/main/ip-adapter.bin"
 file_path = "ip-adapter.bin"
 
-# Check if the file already exists
 if not os.path.exists(file_path):
     print("File not found, downloading...")
     response = requests.get(url, stream=True)
@@ -13,32 +22,23 @@ if not os.path.exists(file_path):
             if chunk:
                 file.write(chunk)
     print("Download completed!")
-else:
-    print("File already exists.")
-
-from models.transformer_sd3 import SD3Transformer2DModel
-import gradio as gr
-import torch
-from pipeline_stable_diffusion_3_ipa import StableDiffusion3Pipeline
-import os
-from PIL import Image
-import spaces
-from huggingface_hub import login
-from diffusers.utils import load_image
 
+# Hugging Face login
 token = os.getenv("HF_TOKEN")
 login(token=token)
 
+# Model paths
 model_path = 'stabilityai/stable-diffusion-3.5-large'
 ip_adapter_path = './ip-adapter.bin'
 image_encoder_path = "google/siglip-so400m-patch14-384"
 
+# Load transformer and pipeline
 transformer = SD3Transformer2DModel.from_pretrained(
-    model_path, subfolder="transformer", torch_dtype=torch.float16
+    model_path, subfolder="transformer", torch_dtype=torch.bfloat16
 )
 
 pipe = StableDiffusion3Pipeline.from_pretrained(
-    model_path, transformer=transformer, torch_dtype=torch.float16
+    model_path, transformer=transformer, torch_dtype=torch.bfloat16
 ).to("cuda")
 
 pipe.init_ipadapter(
@@ -50,10 +50,10 @@ pipe.init_ipadapter(
 
 @spaces.GPU
 def gui_generation(prompt, ref_img, guidance_scale, ipadapter_scale):
-    ref_img = load_image(ref_img.name)
+    # Load and convert reference image
+    ref_img = Image.open(ref_img.name).convert('RGB')
 
     with torch.no_grad():
-        # Ensure the pipeline runs with correct dtype and device
         image = pipe(
             width=1024,
             height=1024,
@@ -62,10 +62,14 @@ def gui_generation(prompt, ref_img, guidance_scale, ipadapter_scale):
             num_inference_steps=24,
             guidance_scale=guidance_scale,
             generator=torch.Generator("cuda").manual_seed(42),
-            clip_image=ref_img.convert('RGB'),
-            ipadapter_scale=ipadapter_scale).images
-    return image[0]
+            clip_image=ref_img,
+            ipadapter_scale=ipadapter_scale
+        ).images[0]
+
+    return image
+
 
+# Set up Gradio interface
 prompt_box = gr.Textbox(label="Prompt", placeholder="Enter your image generation prompt")
 ref_img = gr.File(label="Upload Reference Image")
 guidance_slider = gr.Slider(
@@ -74,7 +78,8 @@ guidance_slider = gr.Slider(
     maximum=16,
     value=7,
     step=0.5,
-    info="Controls adherence to the text prompt")
+    info="Controls adherence to the text prompt"
+)
 
 ipadapter_slider = gr.Slider(
     label="IP-Adapter Scale",
@@ -85,16 +90,12 @@ ipadapter_slider = gr.Slider(
     info="Controls influence of the image prompt"
 )
 
-
-
-# Set up Gradio interface
 interface = gr.Interface(
     fn=gui_generation,
     inputs=[prompt_box, ref_img, guidance_slider, ipadapter_slider],
     outputs="image",
-    title="Image Generation with Stable Diffusion 3 medium and ControlNet",
-    description="Generates an image based on a text prompt and a reference image using Stable Diffusion 3 medium with ControlNet."
-
+    title="Image Generation with Stable Diffusion 3.5 Large and IP-Adapter",
+    description="Generates an image based on a text prompt and a reference image using Stable Diffusion 3.5 Large with IP-Adapter."
 )
 
-interface.launch()
+interface.launch(share=True)