Update app.py

app.py

@@ -6,7 +6,7 @@ from huggingface_hub import hf_hub_download, login
 import random
 import os
 
-# Authentification (gardez ça)
+# Authentification
 hf_token = os.getenv("HF_TOKEN")
 print(f"Token trouvé : {bool(hf_token)}")
 if hf_token:
@@ -23,40 +23,121 @@ def load_lora(repo_id, style):
         filename = f"{style}_lora.safetensors"
         lora_path = hf_hub_download(repo_id=repo_id, filename=filename)
         lora_repo = repo_id
-        return f"✅ LoRA: {repo_id} ({filename})"
+        return f"✅ LoRA chargé : {repo_id} ({filename})"
     except Exception as e:
-        return f"❌ {e}"
+        return f"❌ Erreur : {str(e)}"
 
 @spaces.GPU(duration=120)
-def generate(prompt, negative_prompt, width=1024, height=1024, steps=4, seed=-1, lora_scale=0.8):
-    pipe = FluxPipeline.from_pretrained(model_id, torch_dtype=torch.bfloat16)
-    pipe.to("cuda")
-    
-    if lora_repo and lora_path:
-        pipe.load_lora_weights(lora_path)
-        pipe.fuse_lora(lora_scale=lora_scale)  # Après to("cuda")
+def generate(prompt, negative_prompt="", width=1024, height=1024, steps=4, seed=-1, lora_scale=0.8):
+    try:
+        # Initialisation du pipeline
+        pipe = FluxPipeline.from_pretrained(
+            model_id, 
+            torch_dtype=torch.bfloat16
+        )
+        pipe.to("cuda")
+        
+        # Chargement du LoRA si disponible
+        if lora_repo and lora_path:
+            pipe.load_lora_weights(lora_path)
+            pipe.fuse_lora(lora_scale=lora_scale)
+        
+        # Optimisation mémoire
+        pipe.enable_model_cpu_offload()
+        
+        # Génération du seed
+        if seed == -1:
+            seed = random.randint(0, 2**32 - 1)
+        generator = torch.Generator("cuda").manual_seed(seed)
+        
+        # Génération de l'image
+        image = pipe(
+            prompt=prompt,
+            negative_prompt=negative_prompt if negative_prompt else None,
+            height=height,
+            width=width,
+            num_inference_steps=steps,
+            guidance_scale=0.0,
+            generator=generator,
+            max_sequence_length=256
+        ).images[0]
+        
+        # Nettoyage de la mémoire
+        del pipe
+        torch.cuda.empty_cache()
+        
+        return image
     
-    pipe.enable_model_cpu_offload()  # Optim VRAM
+    except Exception as e:
+        print(f"❌ Erreur de génération : {str(e)}")
+        raise gr.Error(f"Erreur : {str(e)}")
+
+# Interface Gradio
+with gr.Blocks(title="Flux Schnell + LoRA") as demo:
+    gr.Markdown("# 🎨 Générateur Flux Schnell + LoRA")
     
-    generator = torch.Generator("cuda").manual_seed(seed if seed != -1 else random.randint(0, 2**32))
+    with gr.Row():
+        with gr.Column():
+            # LoRA
+            gr.Markdown("### Charger un LoRA (optionnel)")
+            lora_repo_input = gr.Textbox(
+                label="Repository LoRA",
+                placeholder="username/repo-name"
+            )
+            lora_style = gr.Textbox(
+                label="Style LoRA",
+                placeholder="style_name",
+                value="style"
+            )
+            load_btn = gr.Button("Charger LoRA")
+            lora_status = gr.Textbox(label="Status", interactive=False)
+            
+            # Paramètres
+            gr.Markdown("### Paramètres de génération")
+            prompt = gr.Textbox(
+                label="Prompt",
+                placeholder="Décrivez votre image...",
+                lines=3
+            )
+            negative_prompt = gr.Textbox(
+                label="Negative Prompt (optionnel)",
+                placeholder="Ce que vous ne voulez pas...",
+                lines=2
+            )
+            
+            with gr.Row():
+                width = gr.Slider(512, 2048, 1024, step=64, label="Largeur")
+                height = gr.Slider(512, 2048, 1024, step=64, label="Hauteur")
+            
+            with gr.Row():
+                steps = gr.Slider(1, 10, 4, step=1, label="Steps")
+                seed = gr.Number(label="Seed (-1 = aléatoire)", value=-1)
+                lora_scale = gr.Slider(0, 1, 0.8, step=0.1, label="LoRA Scale")
+            
+            generate_btn = gr.Button("🚀 Générer", variant="primary")
+        
+        with gr.Column():
+            output_image = gr.Image(label="Image générée", type="pil")
     
-    image = pipe(
-        prompt, 
-        negative_prompt=negative_prompt,
-        height=height, width=width, 
-        num_inference_steps=steps,
-        guidance_scale=0.0, 
-        generator=generator,
-        max_sequence_length=256  # Pour FLUX
-    ).images[0]
+    # Actions
+    load_btn.click(
+        fn=load_lora,
+        inputs=[lora_repo_input, lora_style],
+        outputs=lora_status
+    )
     
-    del pipe  # Nettoyage
-    torch.cuda.empty_cache()
-    return image
-
-# Interface Gradio (inchangée)
-with gr.Blocks(title="Flux Schnell + LoRA") as demo:
-    # ... (votre interface reste identique)
-    pass
+    generate_btn.click(
+        fn=generate,
+        inputs=[prompt, negative_prompt, width, height, steps, seed, lora_scale],
+        outputs=output_image
+    )
 
-demo.launch()
+# Lancement avec gestion propre de l'event loop
+if __name__ == "__main__":
+    demo.queue()  # Active la file d'attente
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        show_error=True,
+        quiet=False
+    )