only for Image to Text

app.py

@@ -175,7 +175,7 @@ def _ensure_db_ready():
         raise AppError(db_init_error, 503)
 
 
-def _finalize_caption(raw_text: str) -> str:
+def _finalize_caption(raw_text: str, max_sentences: int = CAPTION_MAX_SENTENCES) -> str:
     text = " ".join(raw_text.split()).strip()
     if not text:
         return ""
@@ -184,7 +184,7 @@ def _finalize_caption(raw_text: str) -> str:
     sentences = [s.strip() for s in sentences if s.strip()]
 
     if len(sentences) >= CAPTION_MIN_SENTENCES:
-        return " ".join(sentences[:CAPTION_MAX_SENTENCES]).strip()
+        return " ".join(sentences[:max_sentences]).strip()
 
     if text and text[-1] not in ".!?":
         text = re.sub(r"[,:;\-]\s*[^,:;\-]*$", "", text).strip()
@@ -261,8 +261,8 @@ def summarize_captions(captions: list[str]) -> str:
         with torch.no_grad():
             output_ids = model.generate(
                 **inputs,
-                max_length=150,
-                min_length=20,
+                max_length=300,
+                min_length=50,
                 length_penalty=2.0,
                 num_beams=4,
                 early_stopping=True,
@@ -271,7 +271,7 @@ def summarize_captions(captions: list[str]) -> str:
     except Exception as exc:
         raise AppError("Failed to summarize captions.", 500) from exc
 
-    return _finalize_caption(summary)
+    return _finalize_caption(summary, max_sentences=10)
 
 
 def generate_caption_text(image: Image.Image) -> str: