diff --git a/management/commands/generate_clips.py b/management/commands/generate_clips.py
index 4922d53..3d0091e 100644
--- a/management/commands/generate_clips.py
+++ b/management/commands/generate_clips.py
@@ -19,10 +19,12 @@ class Command(BaseCommand):
         clips = []
         for i in item.models.Item.objects.filter(data__type__contains="Original"):
             qs = item.models.Item.objects.filter(data__title=i.data['title']).exclude(id=i.id)
-            if qs.count() == 2:
+            if qs.count() >= 2:
                 clip = {}
                 durations = []
                 for e in item.models.Item.objects.filter(data__title=i.data['title']):
+                    if 'type' not in e.data:
+                        print("ignoring invalid video", e)
                     source = e.files.all()[0].data.path
                     ext = os.path.splitext(source)[1]
                     type_ = e.data['type'][0].lower()
@@ -35,7 +37,10 @@ class Command(BaseCommand):
                     durations.append(e.files.all()[0].duration)
                 clip["duration"] = min(durations)
                 clip['tags'] = i.data.get('tags', [])
-                clips.append(clip)
+                if "original" in clip and "foreground" in clip and "background" in clip:
+                    clips.append(clip)
+                else:
+                    print("ignoring incomplete video", i)
 
         with open(os.path.join(prefix, 'clips.json'), 'w') as fd:
             json.dump(clips, fd, indent=2, ensure_ascii=False)
diff --git a/management/commands/render.py b/management/commands/render.py
index b43dbc9..24ba964 100644
--- a/management/commands/render.py
+++ b/management/commands/render.py
@@ -1,5 +1,6 @@
 import json
 import os
+import subprocess
 
 from django.core.management.base import BaseCommand
 from django.conf import settings
@@ -14,6 +15,7 @@ class Command(BaseCommand):
         parser.add_argument('--prefix', action='store', dest='prefix', default="/srv/t_for_time", help='prefix to build clips in')
         parser.add_argument('--duration', action='store', dest='duration', default="150", help='target duration in seconds')
         parser.add_argument('--offset', action='store', dest='offset', default="1024", help='inital offset in pi')
+        parser.add_argument('--no-video', action='store_true', dest='no_video', default=False, help='don\'t render video')
 
     def handle(self, **options):
         prefix = options['prefix']
@@ -23,7 +25,15 @@ class Command(BaseCommand):
         with open(os.path.join(prefix, "clips.json")) as fd:
             clips = json.load(fd)
         scene = compose(clips, target=target, base=base)
-        render(prefix, scene, 'scene-%s-' % base)
+        timelines = render(prefix, scene, 'scene-%s-' % base)
         with open(os.path.join(prefix, 'scene-%s.json' % base), 'w') as fd:
             json.dump(scene, fd, indent=2, ensure_ascii=False)
+        if not options['no_video']:
+            for timeline in timelines:
+                cmd = [
+                    'xvfb-run', '-a',
+                    'melt', timeline,
+                    '-consumer', 'avformat:%s' % timeline.replace('.kdenlive', '.mp4')
+                ]
+                subprocess.call(cmd)
 
diff --git a/render.py b/render.py
index cdb70eb..0a879be 100644
--- a/render.py
+++ b/render.py
@@ -14,7 +14,9 @@ def random_choice(seq, items, pop=False):
     n = n_ = len(items) - 1
     #print('len', n)
     if n == 0:
-        return items[0]
+        if pop:
+            return items.pop(n)
+        return items[n]
     r = seq()
     base = 10
     while n > 10:
@@ -50,9 +52,12 @@ def compose(clips, target=150, base=1024):
             'A4': [],
         }
     }
+    all_clips = clips.copy()
     seq = random(base)
     while target - length > 10 and clips:
         clip = random_choice(seq, clips, True)
+        if not clips:
+            clips = [c for c in all_clips if c != clip]
         if length + clip['duration'] > target:
             break
         length += clip['duration']
@@ -112,6 +117,7 @@ def compose(clips, target=150, base=1024):
 
 def render(root, scene, prefix=''):
     fps = 24
+    files = []
     for timeline, data in scene.items():
         print(timeline)
         project = KDEnliveProject(root)
@@ -121,6 +127,8 @@ def render(root, scene, prefix=''):
             print(track)
             for clip in clips:
                 project.append_clip(track, clip)
-
-        with open(os.path.join(root, prefix + "%s.kdenlive" % timeline), 'w') as fd:
+        path = os.path.join(root, prefix + "%s.kdenlive" % timeline)
+        with open(path, 'w') as fd:
             fd.write(project.to_xml())
+        files.append(path)
+    return files
diff --git a/render_kdenlive.py b/render_kdenlive.py
index 6daf309..58e68a6 100644
--- a/render_kdenlive.py
+++ b/render_kdenlive.py
@@ -5,6 +5,7 @@ import lxml.etree
 import uuid
 import os
 
+_CACHE = {}
 _IDS = defaultdict(int)
 
 def get_propery(element, name):
@@ -12,6 +13,7 @@ def get_propery(element, name):
 
 
 
+
 class KDEnliveProject:
 
     def to_xml(self):
@@ -316,7 +318,10 @@ class KDEnliveProject:
         return prefix + self.get_counter(prefix)
 
     def get_chain(self, file, kdenlive_id=None):
-        out = subprocess.check_output(['melt', file, '-consumer', 'xml'])
+        if file in _CACHE:
+            out = _CACHE[file]
+        else:
+            out = _CACHE[file] = subprocess.check_output(['melt', file, '-consumer', 'xml'])
         chain = lxml.etree.fromstring(out).xpath('producer')[0]
         chain.tag = 'chain'
         chain.attrib['id'] = self.get_id('chain')