diff --git a/manim_voiceover/modify_audio.py b/manim_voiceover/modify_audio.py
index e0912c6..be57925 100644
--- a/manim_voiceover/modify_audio.py
+++ b/manim_voiceover/modify_audio.py
@@ -2,6 +2,7 @@
 import sox
 import uuid
 from mutagen.mp3 import MP3
+from mutagen.wave import WAVE
 
 
 def adjust_speed(input_path: str, output_path: str, tempo: float) -> None:
@@ -19,6 +20,7 @@ def adjust_speed(input_path: str, output_path: str, tempo: float) -> None:
 
 
 def get_duration(path: str) -> float:
-    audio = MP3(path)
-    return audio.info.length
+    if str(path).endswith(".wav"):
+        return WAVE(path).info.length
+    return MP3(path).info.length
     # return sox.file_info.duration(path)
diff --git a/manim_voiceover/services/gemini.py b/manim_voiceover/services/gemini.py
new file mode 100644
index 0000000..3bd5046
--- /dev/null
+++ b/manim_voiceover/services/gemini.py
@@ -0,0 +1,120 @@
+import os
+import sys
+import wave
+from pathlib import Path
+from dotenv import find_dotenv, load_dotenv
+from manim import logger
+
+from manim_voiceover.helper import (
+    create_dotenv_file,
+    remove_bookmarks,
+)
+from manim_voiceover.services.base import SpeechService
+
+try:
+    from google import genai
+    from google.genai import types
+except ImportError:
+    logger.error(
+        "Missing packages. "
+        'Run `pip install google-genai` to use GeminiTTSService.'
+    )
+
+load_dotenv(find_dotenv(usecwd=True))
+
+
+def create_dotenv_gemini():
+    logger.info("You need a Gemini API key from https://makersuite.google.com/app/apikey")
+    if not create_dotenv_file(["GOOGLE_API_KEY"]):
+        raise ValueError(
+            "The environment variable GOOGLE_API_KEY is not set. "
+            "Please add it to your .env file."
+        )
+    logger.info("The .env file has been created. Please restart Manim.")
+    sys.exit()
+
+
+class GeminiTTSService(SpeechService):
+    """
+    Gemini-based TTS service using Google's Gemini 2.5 SDK.
+    """
+
+    def __init__(
+        self,
+        model: str = "gemini-2.5-flash-preview-tts",
+        voice_name: str = "Kore",
+        **kwargs
+    ):
+        self.model = model
+        self.voice_name = voice_name
+
+        api_key = os.getenv("GOOGLE_API_KEY")
+        if not api_key:
+            create_dotenv_gemini()
+
+        self.client = genai.Client(api_key=api_key)
+
+        super().__init__(transcription_model=None, **kwargs)
+
+
+    def generate_from_text(
+        self,
+        text: str,
+        cache_dir: str = None,
+        path: str = None,
+        **kwargs
+    ) -> dict:
+        if cache_dir is None:
+            cache_dir = self.cache_dir
+
+        clean_text = remove_bookmarks(text)
+
+        input_data = {
+            "input_text": clean_text,
+            "service": "gemini",
+            "config": {
+                "model": self.model,
+                "voice": self.voice_name,
+            },
+        }
+
+        cached = self.get_cached_result(input_data, cache_dir)
+        if cached:
+            return cached
+
+        try:
+            response = self.client.models.generate_content(
+                model=self.model,
+                contents=clean_text,
+                config=types.GenerateContentConfig(
+                    response_modalities=["AUDIO"],
+                    speech_config=types.SpeechConfig(
+                        voice_config=types.VoiceConfig(
+                            prebuilt_voice_config=types.PrebuiltVoiceConfig(
+                                voice_name=self.voice_name,
+                            )
+                        )
+                    ),
+                ),
+            )
+        except Exception as e:
+            logger.error(f"Gemini TTS generation failed: {e}")
+            raise
+
+        audio_data = response.candidates[0].content.parts[0].inline_data.data
+
+        audio_path = path or self.get_audio_basename(input_data) + ".wav"
+        full_path = Path(cache_dir) / audio_path
+
+        # Save audio
+        with wave.open(str(full_path), "wb") as wf:
+            wf.setnchannels(1)
+            wf.setsampwidth(2)
+            wf.setframerate(24000)
+            wf.writeframes(audio_data)
+
+        return {
+            "input_text": text,
+            "input_data": input_data,
+            "original_audio": audio_path,
+        }