Fix image detection and LoRA loading to match training notebook

app.py: check the known save path directly instead of parsing stdout
(broken after PR #21 removed the print statement).

card_generator_adapter.py: two mismatches with the training notebook:
1. LoRA loading used pipe.load_lora_weights() (diffusers format) but the
   adapter was saved with PEFT's save_pretrained() — keys didn't match.
   Now uses PeftModel.from_pretrained() + merge_and_unload().
2. Conditioning built a natural language prompt, but the LoRA was trained
   on json.dumps(meta). Now uses JSON serialization to match.

app.py

@@ -37,6 +37,7 @@ def _extract_image_from_stdout(stdout: str) -> Path | None:
 
 
 def run_prompt_pipeline(prompt_text: str) -> tuple[Path | None, str, list[str]]:
+    save_path = "generated_card.png"
     cmd = [
         sys.executable, "prompt_to_card_pipeline.py",
         prompt_text,
@@ -46,7 +47,7 @@ def run_prompt_pipeline(prompt_text: str) -> tuple[Path | None, str, list[str]]:
         "--template", "clean-text-to-keywords/json_template_example.json",
         "--generator-module", "card_generator_adapter.py",
         "--device", "cuda",
-        "--save-path", "generated_card.png",
+        "--save-path", save_path,
         "--print-json",
     ]
 
@@ -63,7 +64,9 @@ def run_prompt_pipeline(prompt_text: str) -> tuple[Path | None, str, list[str]]:
     if result.returncode != 0:
         return None, full_output.strip() or "Erreur inconnue pendant le pipeline.", cmd
 
-    image_path = _extract_image_from_stdout(result.stdout or "")
+    image_path = APP_DIR / save_path
+    if not image_path.exists():
+        image_path = _extract_image_from_stdout(result.stdout or "")
     return image_path, full_output.strip(), cmd
 
 # ------------------------------------------------------------------ #