diff --git a/encode.py b/encode.py
new file mode 100755
index 0000000..8d56d83
--- /dev/null
+++ b/encode.py
@@ -0,0 +1,24 @@
+#!/usr/bin/python3
+import subprocess
+import os
+from glob import glob
+
+for xml in glob('output/*.xml'):
+    mp4 = xml.replace('.xml', '.mp4')
+    if not os.path.exists(mp4) or os.path.getmtime(xml) > os.path.getmtime(mp4):
+        subprocess.call([
+            'qmelt', xml, '-consumer', 'avformat:' + mp4 + '.mp4', 'vcodec=libx264'
+        ])
+        subprocess.call([
+            'qmelt', xml, '-consumer', 'avformat:' + mp4 + '.wav',
+        ])
+        subprocess.call([
+            'ffmpeg', '-y',
+            '-i', mp4 + '.mp4',
+            '-i', mp4 + '.wav',
+            '-c:v', 'copy',
+            '-map', '0:v',
+            '-map', '1:a',
+            '-strict', '-2',
+            mp4
+        ])
diff --git a/render.py b/render.py
index c4c8624..4f1bb0c 100755
--- a/render.py
+++ b/render.py
@@ -3,6 +3,7 @@ import os
 import sys
 import json
 import subprocess
+from collections import defaultdict
 import string
 from glob import glob
 
@@ -14,6 +15,8 @@ import ox.web.auth
 
 base_url = 'https://cdosea.0x2620.org'
 
+FRAME_DURATION = 1/60
+
 api = None
 
 def get_api():
@@ -32,6 +35,34 @@ if os.path.exists('CLIPS.json'):
 else:
     CLIPS = {}
 
+if not os.path.exists('MUSIC.json'):
+    MUSIC = defaultdict(list)
+    for letter in os.listdir('music'):
+        for d in range(10):
+            path = os.path.join('music', letter, '%d.mp3' % d)
+            MUSIC[letter].append({
+                'path': path,
+                'duration': ox.avinfo(path)['duration']
+            })
+    with open('MUSIC.json', 'w') as fd:
+        json.dump(MUSIC, fd, indent=2)
+else:
+    MUSIC = json.load(open('MUSIC.json'))
+
+if not os.path.exists('VOCALS.json'):
+    VOCALS = defaultdict(list)
+    for letter in os.listdir('vocals'):
+        for fn in sorted(os.listdir(os.path.join('vocals', letter))):
+            path = os.path.join('vocals', letter, fn)
+            VOCALS[letter].append({
+                'path': path,
+                'duration': ox.avinfo(path)['duration']
+            })
+    with open('VOCALS.json', 'w') as fd:
+        json.dump(VOCALS, fd, indent=2)
+else:
+    VOCALS = json.load(open('VOCALS.json'))
+
 def get_path(id):
     global PATHS
     if id not in PATHS:
@@ -81,6 +112,9 @@ def get_clips(tag):
             'range': [0, 10000]})['data']['items']
         for clip in clips:
             clip['path'] = get_path(clip['id'].split('/')[0])
+            # or use round?
+            clip['in'] = int(clip['in'] / FRAME_DURATION) * FRAME_DURATION
+            clip['out'] = int(clip['out'] / FRAME_DURATION) * FRAME_DURATION
             clip['duration'] = clip['out'] - clip['in']
             clip['tag'] = tag
         CLIPS[tag] = list(sorted(clips, key=lambda c: c['id']))
@@ -147,6 +181,8 @@ def sequence(seq, letter):
     result = {
         'clips': [],
         'text': [],
+        'vocals': [],
+        'music': [],
     }
     duration = 0
     MAX_DURATION = 65 * 2
@@ -175,6 +211,8 @@ def sequence(seq, letter):
     for clip in result['clips']:
         if seq() == 0:
             clip['black'] = True
+
+    # text overlay
     position = last_text = 0
     tags_text = []
     while position < duration:
@@ -203,6 +241,45 @@ def sequence(seq, letter):
         blank = {'blank': True, 'duration': duration - last_text}
         result['text'].append(blank)
 
+    # music
+    if letter in MUSIC:
+        position = last_music = 0
+        while position < duration:
+            n = seq()
+            if n == 0:
+                blank = {'blank': True, 'duration': position - last_music}
+                result['music'].append(blank)
+                n = seq()
+
+                clip = MUSIC[letter][n]
+                position += clip['duration']
+                if position > duration and result['music'][-1].get('blank'):
+                    result['music'][-1]['duration'] -= (position-duration)
+                    position = duration
+                result['music'].append(clip)
+                last_music = position
+            else:
+                position += n
+        if last_music < duration:
+            blank = {'blank': True, 'duration': duration - last_music}
+            result['music'].append(blank)
+    # vocals
+    if letter in VOCALS:
+        n = seq()
+        clip = VOCALS[letter][n]
+
+        n = 1.0 / (seq() + 1)  # 0.1 - 1
+
+        silence = duration - clip['duration']
+        silence_start = n * silence
+        blank = {'blank': True, 'duration': silence_start}
+        if n != 0:
+            result['vocals'].append(blank)
+        result['vocals'].append(clip)
+        if n != 1:
+            blank = {'blank': True, 'duration': silence - silence_start}
+            result['vocals'].append(blank)
+
     return result
 
 if __name__ == '__main__':