uuu

app.py

@@ -6,6 +6,7 @@ import spaces
 from PIL import Image
 from huggingface_hub import login
 from diffusers.utils import load_image
+from torchvision import transforms
 
 from models.transformer_sd3 import SD3Transformer2DModel
 from pipeline_stable_diffusion_3_ipa import StableDiffusion3Pipeline
@@ -50,9 +51,18 @@ pipe.init_ipadapter(
 
 @spaces.GPU
 def gui_generation(prompt, ref_img, guidance_scale, ipadapter_scale):
-    # Load and convert reference image
+
+    # Load and preprocess the reference image
+    preprocess = transforms.Compose([
+        transforms.Resize((384, 384)),
+        transforms.ToTensor(),
+        transforms.ConvertImageDtype(torch.float16)
+    ])
+
     ref_img = Image.open(ref_img.name).convert('RGB')
+    ref_img_tensor = preprocess(ref_img).unsqueeze(0).to("cuda")
 
+    # Generate the image
     with torch.no_grad():
         image = pipe(
             width=1024,
@@ -62,7 +72,7 @@ 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,
+            clip_image=ref_img_tensor,
             ipadapter_scale=ipadapter_scale
         ).images[0]
 

requirements.txt

@@ -8,4 +8,5 @@ sentencepiece
 spaces
 pillow
 peft
-openai
+openai
+torchvision