Merge branch 'main' of https://huggingface.co/spaces/Hatman/InstantStyle

app.py

@@ -3,20 +3,44 @@ import random
 import spaces
 import gradio as gr
 from PIL import Image
-from diffusers import AutoPipelineForText2Image
-from diffusers.utils import load_image
+from models_transformer_sd3 import SD3Transformer2DModel
+from pipeline_stable_diffusion_3_ipa import StableDiffusion3Pipeline
+import gc
+import os
+from huggingface_hub import login
+
+TOKEN = os.getenv('TOKEN')
+login(TOKEN)
+
+model_path = 'stabilityai/stable-diffusion-3.5-large'
+ip_adapter_path = './ip-adapter.bin'
+image_encoder_path = "google/siglip-so400m-patch14-384"
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 dtype = torch.float16 if torch.cuda.is_available() else torch.float32
-pipe = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=dtype)
-pipe.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter_sdxl.bin")
+
+transformer = SD3Transformer2DModel.from_pretrained(
+    model_path, subfolder="transformer", torch_dtype=torch.bfloat16
+)
+
+pipe = StableDiffusion3Pipeline.from_pretrained(
+    model_path, transformer=transformer, torch_dtype=torch.bfloat16
+) ## For ZeroGPU no .to("cuda")
+
+pipe.init_ipadapter(
+    ip_adapter_path=ip_adapter_path, 
+    image_encoder_path=image_encoder_path, 
+    nb_token=64, 
+)
+
 pipe.to(device)
+
 def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
     if randomize_seed:
         seed = random.randint(0, 2000)
     return seed
-    
-@spaces.GPU()
+
+@spaces.GPU() ## For ZeroGPU
 def create_image(image_pil,
                  prompt,
                  n_prompt,
@@ -28,6 +52,9 @@ def create_image(image_pil,
                  target="Load only style blocks",
                  ):
     
+    if image_pil is None:
+        return None
+    
     if target !="Load original IP-Adapter":
         if target=="Load only style blocks":
             scale = {
@@ -43,19 +70,25 @@ def create_image(image_pil,
                 "up": {"block_0": [0.0, control_scale, 0.0]},
             }
         pipe.set_ip_adapter_scale(scale)
-
-    style_image = load_image(image_pil)
-    generator = torch.Generator().manual_seed(randomize_seed_fn(seed, True))
+    
+    style_image = Image.open(image_pil).convert('RGB')
     
     
     image = pipe(
-        prompt=prompt,
-        ip_adapter_image=style_image,
-        negative_prompt=n_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        generator=generator,
-    ).images[0]
+            width=1024,
+            height=1024,
+            prompt=prompt,
+            negative_prompt="lowres, low quality, worst quality",
+            num_inference_steps=24, 
+            guidance_scale=guidance_scale,
+            generator=torch.Generator("cuda").manual_seed(randomize_seed_fn(seed, True)),  ## For ZeroGPU no device="cpu"
+            clip_image=style_image,
+            ipadapter_scale=scale,
+        ).images[0]
+    
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+        gc.collect()
     
     return image