Index: libhb/muxmp4.c
===================================================================
--- libhb/muxmp4.c	(revision 1128)
+++ libhb/muxmp4.c	(working copy)
@@ -412,16 +412,6 @@
         duration = MP4_INVALID_DURATION;
     }
 
-    /* When we do get the first keyframe, use its duration as the
-       initial delay added to the frame order offset for b-frames.
-       Because of b-pyramid, double this duration when there are
-       b-pyramids, as denoted by job->areBframes equalling 2. */
-    if ((mux_data->track == 1) && (thisSample == 0) && (buf->frametype & HB_FRAME_KEY) && (job->vcodec == HB_VCODEC_X264))
-    {
-        initDelay = buf->renderOffset;
-        thisSample++;
-    }
-
     /* Here's where the sample actually gets muxed. 
        If it's an audio sample, don't offset the sample's playback.
        If it's a video sample and there are no b-frames, ditto.
@@ -474,21 +464,6 @@
            MP4SampleId i;
            MP4Duration renderingOffset = 2000000000, tmp;
            
-           // Find the smallest rendering offset
-           for(i = 1; i <= count; i++)
-           {
-               tmp = MP4GetSampleRenderingOffset(m->file, 1, i);
-               if(tmp < renderingOffset)
-                   renderingOffset = tmp;
-           }
-           
-           // Adjust all ctts values down by renderingOffset
-           for(i = 1; i <= count; i++)
-           {
-               MP4SetSampleRenderingOffset(m->file,1,i,
-                   MP4GetSampleRenderingOffset(m->file,1,i) - renderingOffset);
-           }
-           
            // Insert track edit to get A/V back in sync.  The edit amount is
            // the rendering offset of the first sample.
            MP4AddTrackEdit(m->file, 1, MP4_INVALID_EDIT_ID, MP4GetSampleRenderingOffset(m->file,1,1),
Index: libhb/encx264.c
===================================================================
--- libhb/encx264.c	(revision 1128)
+++ libhb/encx264.c	(working copy)
@@ -35,6 +35,7 @@
     // Internal queue of DTS start/stop values.
     int64_t        dts_start[DTS_BUFFER_SIZE];
     int64_t        dts_stop[DTS_BUFFER_SIZE];
+    int64_t        init_delay;
 
     int64_t        dts_write_index;
     int64_t        dts_read_index;
@@ -250,6 +251,32 @@
     pv->dts_read_index = 0;
     pv->next_chap = 0;
 
+    if (job->areBframes)
+    {
+        /* Basic initDelay value is the clockrate divided by the FPS
+           -- the length of one frame in clockticks.                  */
+        pv->init_delay = (float)90000 / (float)((float)job->vrate / (float)job->vrate_base);
+        
+        /* 23.976-length frames are 3753.75 ticks long. That means 25%
+           will come out as 3753, 75% will be 3754. The delay has to be
+           the longest possible frame duration, 3754. However, 3753.75
+           gets truncated to 3753, so if that's what it is, ++ it.     */
+        if (pv->init_delay == 3753)
+            pv->init_delay++;
+        
+        /* For VFR, libhb sees the FPS as 29.97, but the longest frames
+           will use the duration of frames running at 23.976fps instead.. */
+        if (job->vfr)
+        {
+            pv->init_delay = 3754;
+        }
+    
+        /* The delay is 2 frames for regular b-frames, 3 for b-pyramid.
+           Since job->areBframes is 1 for b-frames and 2 for b-pyramid,
+           add one to it and use it as a multiplier.                    */
+        pv->init_delay *= ( job->areBframes + 1); 
+    }
+    
     return 0;
 }
 
@@ -447,7 +474,7 @@
                        value is pretty much guaranteed to be positive.  The
                        muxing code will minimize the renderOffset at the end. */
 
-                    buf->renderOffset = pic_out.i_pts - dts_start + 1000000;
+                    buf->renderOffset = pic_out.i_pts - dts_start + pv->init_delay;
 
                     /* Send out the next dts values */
                     buf->start = dts_start;