diff --git a/generate.py b/generate.py
index a88a764..9dff58d 100644
--- a/generate.py
+++ b/generate.py
@@ -575,11 +575,15 @@ def replace_character_motion_control(item, character, keep=False):
     return ai
 
 
-def describe_video(url, neutral=False):
+def describe_video(url, neutral=False, face=False):
+    if face:
+        extra = 'describe the facial expression to allow an actor to recreate the scene, '
+    else:
+        extra = ''
     if neutral:
         prompt = (
             "Detect cuts or scene changes and describe each scene, use as much details as you can. "
-            "Describe each person incudling detalied apreance, haircut in a gender neutral way, "
+            f"Describe each person incudling detalied apreance, haircut in a gender neutral way, {extra}"
             "describe each objects, animal or plant, describe foreground and backgroud, "
             "describe from what angle the scene is filmed, incude details about camera model, lense, depth of field used to film this scene. "
             "Use the format: <description of scene 1>. CAMERA CUT TO <description of scene 2>. CAMERA CUT TO <description of scene 3>. "
@@ -616,15 +620,15 @@ def describe_item(item, neutral=False):
     return describe_video(video_url, neutral)
 
 
-def reshoot_item(item, extra_prompt=None, first_frame=None, keep=False):
+def reshoot_item(item, extra_prompt=None, first_frame=None, keep=False, prompt=None):
     if isinstance(item, str):
         item = Item.objects.get(public_id=item)
     if isinstance(first_frame, Document):
         first_frame = public_document_url(first_frame)
     duration = item.sort.duration
     frames = int(duration * 24)
-    prompt = describe_item(item, first_frame is not None)
-
+    if prompt is None:
+        prompt = describe_item(item, first_frame is not None)
     if extra_prompt:
         prompt += " " + extra_prompt
     prompt_hash = hashlib.sha1((prompt).encode()).hexdigest()
@@ -719,8 +723,6 @@ def reshoot_item_segments(item, character, keep=False):
     ai.data["model"] = status["model"]
     ai.data["seed"] = seed
     ai.save()
-    if not keep:
-        shutil.rmtree(os.path.dirname(joined_output))
     for first_frame in first_frames:
         first_frame.add(ai)
     if not keep:
@@ -1146,6 +1148,7 @@ def process_motion_firstframe(character="P1", keep=False):
             try:
                 replace_character_motion_control(i, character, keep=keep)
             except:
+                i.refresh_from_db()
                 add_tag(i, 'ai-failed')
                 print('>> failed', i)
 
@@ -1159,10 +1162,31 @@ def extract_firstframe(character='P1'):
             try:
                 first_frame = replace_character(item, character, 0)
             except:
+                item.refresh_from_db()
                 add_tag(item, 'ai-failed')
 
 def process_reshoot_firstframe():
     l = itemlist.models.List.objects.get(name='Reshoot-Firstframe')
+    for i in l.items.all():
+        if i.sort.duration > 30: continue
+        if i.public_id == 'HZ': continue
+        if i.documents.all().count():
+            ai = Item.objects.filter(data__type__icontains='ai').filter(data__title=i.data['title'])
+            if ai.exists() or 'ai-failed' in i.data.get('tags', []):
+                print('>> skip', i)
+                continue
+            first_frame = i.documents.all().order_by('-created').first()
+            if not first_frame:
+                first_frame = replace_character(i, 'P1', 0)
+            print(i, first_frame, i.documents.all().count())
+            try:
+                reshoot_item(i, first_frame=first_frame)
+            except:
+                add_tag(i, 'ai-failed')
+                print('>> failed', i)
+
+def process_motion_firstframe():
+    l = itemlist.models.List.objects.get(name='Motion-Firstframe')
     for i in l.items.all():
         if i.sort.duration > 30: continue
         if i.public_id == 'HZ': continue