Update app.py

app.py

@@ -1,5 +1,3 @@
-
-
 ##############################
 # ===== Standard Imports =====
 ##############################
@@ -10,12 +8,21 @@ import random
 import json
 from math import floor
 from typing import Any, Dict, List, Optional, Union
-from flux_app.lora import loras
+
+# Local import for default LoRA list (if available)
+try:
+    from flux_app.lora import loras
+except ImportError:
+    loras = [
+        {"image": "placeholder.jpg", "title": "Placeholder LoRA", "repo": "placeholder/repo", "weights": None, "trigger_word": ""}
+    ]
+
 import torch
 import numpy as np
 import requests
 from PIL import Image
 import spaces
+
 # Diffusers imports
 from diffusers import (
     DiffusionPipeline,
@@ -28,14 +35,12 @@ from diffusers.utils import load_image
 # Hugging Face Hub
 from huggingface_hub import ModelCard, HfFileSystem
 
-
 # Gradio (UI)
 import gradio as gr
 
 ##############################
 # ===== config.py =====
 ##############################
-# Configuration parameters
 DTYPE = torch.bfloat16
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 BASE_MODEL = "black-forest-labs/FLUX.1-dev"
@@ -167,7 +172,7 @@ def generate(message, max_new_tokens=256, temperature=0.9, top_p=0.95, repetitio
 ##############################
 # ===== lora_handling.py =====
 ##############################
-# A default list of LoRAs for the UI (this would normally be loaded from a separate module)
+# A default list of LoRAs for the UI
 loras = [
     {"image": "placeholder.jpg", "title": "Placeholder LoRA", "repo": "placeholder/repo", "weights": None, "trigger_word": ""}
 ]
@@ -446,7 +451,7 @@ class ModelManager:
         """Initializes the diffusion pipelines and autoencoders."""
         self.taef1 = AutoencoderTiny.from_pretrained(TAEF1_MODEL, torch_dtype=DTYPE).to(DEVICE)
         self.good_vae = AutoencoderKL.from_pretrained(BASE_MODEL, subfolder="vae", torch_dtype=DTYPE).to(DEVICE)
-        # Optionally, if your model is private, you can pass `use_auth_token=self.hf_token` here.
+        # Optionally, pass use_auth_token=self.hf_token if needed.
         self.pipe = DiffusionPipeline.from_pretrained(BASE_MODEL, torch_dtype=DTYPE, vae=self.taef1)
         self.pipe = self.pipe.to(DEVICE)
         self.pipe_i2i = AutoPipelineForImage2Image.from_pretrained(
@@ -459,8 +464,9 @@ class ModelManager:
             tokenizer_2=self.pipe.tokenizer_2,
             torch_dtype=DTYPE,
         ).to(DEVICE)
-        # Bind the custom LoRA call to the pipeline.
-        self.pipe.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images.__get__(self.pipe)
+        # Instead of binding to the instance (which fails due to __slots__),
+        # bind the custom method to the pipeline’s class.
+        self.pipe.__class__.flux_pipe_call_that_returns_an_iterable_of_images = flux_pipe_call_that_returns_an_iterable_of_images
     
     @spaces.GPU(duration=100)
     def generate_image(self, prompt_mash, steps, seed, cfg_scale, width, height, lora_scale):
@@ -504,8 +510,6 @@ class ModelManager:
 ##############################
 # ===== frontend.py =====
 ##############################
-
-
 class Frontend:
     def __init__(self, model_manager: ModelManager):
         self.model_manager = model_manager
@@ -722,4 +726,5 @@ if __name__ == "__main__":
     frontend = Frontend(model_manager)
     app = frontend.create_ui()
     app.queue()
-    app.launch()
+    # Set share=True to create a public link if desired.
+    app.launch(share=False, debug=True)