diff --git a/.env.example b/.env.example
index beec715..b875020 100644
--- a/.env.example
+++ b/.env.example
@@ -6,6 +6,9 @@
 LOG_LEVEL=INFO
 
 TTS_BACKEND=f5_tts
+# F5-TTS model name. Built-in options: F5TTS_v1_Base, F5TTS_v1_Small.
+# Downloaded automatically on first use via HuggingFace.
+TTS_MODEL_NAME=F5TTS_v1_Base
 # TTS_MODEL_PATH=models/f5-tts/model.pt
 # TTS_VOCAB_PATH=models/f5-tts/vocab.txt
 TTS_SAMPLE_RATE=24000
@@ -13,9 +16,9 @@
 VOICES_DIR=voices
 # Path to default reference audio (relative to project root or absolute).
 # Providing DEFAULT_VOICE_REF enables instant warm-up and voice cloning.
-# DEFAULT_VOICE_REF=voices/rick_ref_clean.wav
+DEFAULT_VOICE_REF=voices/rick_ref_clean.wav
 # Exact transcript of the reference audio. When set, Whisper transcription is skipped.
-# DEFAULT_REF_TEXT=Ва-ба-ла-ба-дап-дап! Рикки-тики-тави, сученька! И вот такие у нас новости! Иди.
+DEFAULT_REF_TEXT="Ва-ба-ла-ба-дап-дап! Рикки-тики-тави, сученька! И вот такие у нас новости! Иди."
 
 MIN_SEGMENT_LENGTH=30
 MAX_SEGMENT_LENGTH=200
diff --git a/src/voice_tts/api/server.py b/src/voice_tts/api/server.py
index b0b62cc..1c27a71 100644
--- a/src/voice_tts/api/server.py
+++ b/src/voice_tts/api/server.py
@@ -336,7 +336,13 @@
             f"Unknown TTS backend: {backend}. "
             f"Available backends: {list(_BACKEND_MAP.keys())}"
         )
-    engine = engine_cls(sample_rate=settings.tts_sample_rate)
+    if engine_cls is F5TTSEngine:
+        engine = engine_cls(
+            model=settings.tts_model_name,
+            sample_rate=settings.tts_sample_rate,
+        )
+    else:
+        engine = engine_cls(sample_rate=settings.tts_sample_rate)
     if hasattr(engine, "load"):
         engine.load()
     return engine
diff --git a/src/voice_tts/config.py b/src/voice_tts/config.py
index 57dedcf..00d3788 100644
--- a/src/voice_tts/config.py
+++ b/src/voice_tts/config.py
@@ -11,6 +11,7 @@
 
     # TTS model configuration
     tts_backend: str = "f5_tts"  # or "dummy" for tests
+    tts_model_name: str = "F5TTS_v1_Base"  # env: TTS_MODEL_NAME
     tts_model_path: Path | None = None
     tts_vocab_path: Path | None = None
     tts_sample_rate: int = 24_000