Install required modules

__init__.py

@@ -0,0 +1 @@
+from . import modules

app.py

@@ -4,6 +4,11 @@ from PIL import Image
 import os
 import gradio as gr
 from huggingface_hub import hf_hub_download, snapshot_download
+import shlex
+import subprocess
+
+subprocess.run(shlex.split('pip install flash-attn --no-build-isolation'),
+               env=os.environ | {'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"})
 
 def resolve_hf_path(path):
     if isinstance(path, str) and path.startswith("hf://"):
@@ -44,6 +49,7 @@ from inference.registry import INFERENCES
 
 config_path = os.path.join("config", "ace_plus_fft.yaml")
 cfg = Config(load=True, cfg_file=config_path)
+model_cfg = Config(load=True, cfg_file="config/ace_plus_fft.yaml")
 
 # Instantiate the ACEInference object.
 ace_infer = ACEInference(cfg)
@@ -93,17 +99,22 @@ def face_swap_app(target_img, face_img):
     face_img = face_img.convert("RGB")
 
     edit_mask = create_face_mask(face_img)
+    print(edit_mask)
 
     output_img, edit_image, change_image, mask, seed = ace_infer(
         reference_image=target_img,
         edit_image=face_img,
         edit_mask=edit_mask,
-        prompt="Face swap",
+        prompt="{image}, the person faces the camera.",
         output_height=1024,
         output_width=1024,
         sampler='flow_euler',
         sample_steps=28,
         guide_scale=50,
+        repainting_scale=1.0,
+        use_change=True,
+        keep_pixels=True,
+        keep_pixels_rate=0.8,
         seed=-1
     )
     return output_img

modules/layers.py

@@ -18,6 +18,7 @@ try:
     )
     FLASHATTN_IS_AVAILABLE = True
 except ImportError:
+    print("Importing error!")
     FLASHATTN_IS_AVAILABLE = False
     flash_attn_varlen_func = None
 

requirements.txt

@@ -1,10 +1,25 @@
+huggingface_hub
 diffusers
-gradio
-scepter
-torch
+transformers==4.49.0
+torch==2.4.0
+xformers==0.0.27.post2
 torchvision
-transformers
-huggingface_hub
-Pillow
-numpy
-opencv-python
+gradio==4.44.1
+opencv-python==4.10.0.84
+albumentations
+beautifulsoup4
+bitsandbytes
+einops
+imagehash
+modelscope[framework]
+ms-swift
+open-clip-torch
+opencv-transforms>=0.0.6
+oss2>=2.15.0
+pycocotools
+scikit-image
+scikit-learn
+sentencepiece
+tiktoken
+torchsde
+transformers-stream-generator