Update app.py

app.py

@@ -1,50 +1,17 @@
 import os
-import torch
-from diffusers import AutoPipelineForText2Image
-import gradio as gr
+from huggingface_hub import InferenceClient
 
-# Detectar dispositivo
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-# Cargar modelo con token (OBLIGATORIO para FLUX)
-pipe = AutoPipelineForText2Image.from_pretrained(
-    "black-forest-labs/FLUX.1-dev",
-    torch_dtype=torch.float16,  # más compatible que bfloat16
-    token=os.getenv("HF_TOKEN")  # <-- clave aquí
+client = InferenceClient(
+    provider="fal-ai",
+    api_key=os.environ["HF_TOKEN"],
 )
-pipe = pipe.to(device)
-
-# Optimización (MUY recomendado en Spaces)
-pipe.enable_attention_slicing()
-pipe.enable_vae_slicing()
-
-# Opcional: si tienes poca VRAM
-# pipe.enable_model_cpu_offload()
 
-# Cargar LoRA
-try:
-    pipe.load_lora_weights(
-        "enhanceaiteam/Flux-uncensored-v2",
-        weight_name="lora.safetensors"
-    )
-except Exception as e:
-    print("Error cargando LoRA:", e)
-
-# Función para generar imagen
-def generate(prompt):
-    image = pipe(
-        prompt,
-        num_inference_steps=20,  # más rápido
-        guidance_scale=7.5
-    ).images[0]
-    return image
-
-# UI
-demo = gr.Interface(
-    fn=generate,
-    inputs=gr.Textbox(label="Prompt"),
-    outputs=gr.Image(label="Resultado"),
-    title="FLUX + LoRA Demo"
-)
+with open("cat.png", "rb") as image_file:
+   input_image = image_file.read()
 
-demo.launch()
+# output is a PIL.Image object
+image = client.image_to_image(
+    input_image,
+    prompt="Turn the cat into a tiger.",
+    model="ScottzillaSystems/qwen-image-edit-plus-nsfw-lora",
+)