Update app.py

app.py

@@ -1,80 +1,3 @@
-import spaces
-import gradio as gr
-import torch
-from PIL import Image
-from diffusers import DiffusionPipeline
-import random
-import os
-import json
-import io
-import uuid
-from gradio_client import Client as client_gradio
-from supabase import create_client, Client
-from datetime import datetime
-
-# Inicializa Supabase
-url: str = os.getenv('SUPABASE_URL')
-key: str = os.getenv('SUPABASE_KEY')
-supabase: Client = create_client(url, key)
-
-# Obtém token da Hugging Face
-hf_token = os.getenv("HF_TOKEN")
-
-# Inicializa o modelo base FLUX.1-dev
-base_model = "black-forest-labs/FLUX.1-dev"
-pipe = DiffusionPipeline.from_pretrained(
-    base_model, 
-    torch_dtype=torch.float16,  # Pode testar torch.float32 se a GPU permitir
-    use_safetensors=True
-)
-
-# Move o modelo para GPU
-pipe.to("cuda")
-
-# Definição dos LoRA e Trigger Words
-lora_models = {
-    "Paula": {
-        "repo": "vcollos/Paula2",
-        "weights": "Paula P.safetensors",
-        "trigger_word": ""  # Sem trigger word específica
-    },
-    "Vivi": {
-        "repo": "vcollos/Vivi",
-        "weights": "Vivi.safetensors",
-        "trigger_word": ""
-    }
-}
-
-# Carrega os LoRAs disponíveis
-for name, details in lora_models.items():
-    try:
-        pipe.load_lora_weights(details["repo"], weight_name=details["weights"], adapter_name=name)
-        print(f"✅ LoRA {name} carregado")
-    except Exception as e:
-        print(f"❌ Erro ao carregar o LoRA {name}: {e}")
-
-# Define seed máximo
-MAX_SEED = 2**32 - 1
-
-def upload_image_to_supabase(image, filename):
-    """ Faz upload da imagem para o Supabase Storage e retorna a URL pública. """
-    img_bytes = io.BytesIO()
-    image.save(img_bytes, format="PNG")
-    img_bytes.seek(0)  # Move para o início do arquivo
-
-    storage_path = f"images/{filename}"
-
-    try:
-        # Faz upload da imagem para o Supabase
-        supabase.storage.from_("images").upload(storage_path, img_bytes.getvalue(), {"content-type": "image/png"})
-
-        # Retorna a URL pública da imagem
-        base_url = f"{url}/storage/v1/object/public/images"
-        return f"{base_url}/{filename}"
-    except Exception as e:
-        print(f"❌ Erro no upload da imagem: {e}")
-        return None
-
 @spaces.GPU(duration=80)
 def run_lora(prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_option, lora_scale_1, lora_scale_2, progress=gr.Progress(track_tqdm=True)):
     if randomize_seed:
@@ -97,14 +20,12 @@ def run_lora(prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora
         adapter_weights.append(lora_scale_2)
     elif lora_option == "Ambos":
         selected_loras = ["Paula", "Vivi"]
-        adapter_weights = [lora_scale_1, lora_scale_2]
+        # Ajusta os pesos para garantir equilíbrio
+        total_scale = lora_scale_1 + lora_scale_2
+        adapter_weights = [lora_scale_1 / total_scale, lora_scale_2 / total_scale]
 
     pipe.set_adapters(selected_loras, adapter_weights)
 
-    # Adiciona trigger words apenas se Vivi estiver ativado
-    if "Vivi" in selected_loras:
-        prompt = f"{lora_models['Vivi']['trigger_word']} {prompt}"
-
     # Gera a imagem com precisão de 16 bits para tentar melhorar a nitidez
     with torch.autocast("cuda"):
         image = pipe(
@@ -152,35 +73,4 @@ def run_lora(prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora
     except Exception as e:
         print(f"❌ Erro ao salvar metadados no Supabase: {e}")
 
-    return image, seed
-
-# Interface Gradio
-gr_theme = os.getenv("THEME")
-with gr.Blocks(theme=gr_theme) as app:
-    gr.Markdown("# Paula Image Generator")
-    
-    with gr.Row():
-        with gr.Column(scale=2):
-            prompt = gr.TextArea(label="Prompt", placeholder="Digite um prompt (máx 77 caracteres)", lines=3)
-            generate_button = gr.Button("Gerar")
-            cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=3.5)
-            steps = gr.Slider(label="Steps", minimum=1, maximum=100, step=1, value=32)
-            width = gr.Slider(label="Width", minimum=256, maximum=1024, step=64, value=768)  
-            height = gr.Slider(label="Height", minimum=256, maximum=1024, step=64, value=1024)  
-            randomize_seed = gr.Checkbox(False, label="Randomize seed")
-            seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=556215326)
-            lora_option = gr.Radio(["Nenhum", "Paula", "Vivi", "Ambos"], label="Escolha o LoRA", value="Ambos")
-            lora_scale_1 = gr.Slider(label="LoRA Scale (Paula)", minimum=0, maximum=1, step=0.01, value=1)
-            lora_scale_2 = gr.Slider(label="LoRA Scale (Vivi)", minimum=0, maximum=1, step=0.01, value=1)
-
-        with gr.Column(scale=2):
-            result = gr.Image(label="Generated Image")
-
-    generate_button.click(
-        run_lora,
-        inputs=[prompt, cfg_scale, steps, randomize_seed, seed, width, height, lora_option, lora_scale_1, lora_scale_2],
-        outputs=[result, seed],
-    )
-
-app.queue()
-app.launch(share=True)
+    return image, seed