diff --git a/pico/draw.c b/pico/draw.c
index 236e347e..38b920d6 100644
--- a/pico/draw.c
+++ b/pico/draw.c
@@ -1688,13 +1688,13 @@ void FinalizeLine8bit(int sh, int line, struct PicoEState *est)
   int len;
   static int dirty_line;
 
+  // a hack for mid-frame palette changes
   if (Pico.m.dirtyPal == 1)
   {
-    // a hack for mid-frame palette changes
-    if (!(est->rendstatus & PDRAW_SONIC_MODE) | (line - dirty_line > 4)) {
-      // store a maximum of 3 additional palettes in SonicPal
-      if (est->SonicPalCount < 3)
-        est->SonicPalCount ++;
+    // store a maximum of 2 additional palettes in SonicPal
+    if (est->SonicPalCount < 2 &&
+        (!(est->rendstatus & PDRAW_SONIC_MODE) || (line - dirty_line > 4))) {
+      est->SonicPalCount ++;
       dirty_line = line;
       est->rendstatus |= PDRAW_SONIC_MODE;
     }
@@ -1854,7 +1854,6 @@ static int DrawDisplay(int sh)
 PICO_INTERNAL void PicoFrameStart(void)
 {
   int loffs = 8, lines = 224, coffs = 0, columns = 320;
-  int dirty = ((Pico.est.rendstatus & PDRAW_SONIC_MODE) || Pico.m.dirtyPal);
   int sprep = Pico.est.rendstatus & (PDRAW_SPRITES_MOVED|PDRAW_DIRTY_SPRITES);
   int skipped = Pico.est.rendstatus & PDRAW_SKIP_FRAME;
 
@@ -1896,7 +1895,7 @@ PICO_INTERNAL void PicoFrameStart(void)
   if (FinalizeLine == FinalizeLine8bit) {
     // make a backup of the current palette in case Sonic mode is detected later
     Pico.est.SonicPalCount = 0;
-    Pico.m.dirtyPal = (dirty ? 2 : 0); // mark as dirty but already copied
+    Pico.m.dirtyPal = (Pico.m.dirtyPal ? 2 : 0); // mark as dirty but copied
     blockcpy(Pico.est.SonicPal, PicoMem.cram, 0x40*2);
   }
 }
@@ -2011,6 +2010,8 @@ void PicoDrawUpdateHighPal(void)
       blockcpy(est->HighPal+0x40, est->HighPal, 0x40*2);
       blockcpy(est->HighPal+0x80, est->HighPal, 0x80*2);
     }
+    Pico.est.HighPal[0xe0] = 0x0000; // black and white, reserved for OSD
+    Pico.est.HighPal[0xf0] = 0xffff;
   }
 }
 
diff --git a/pico/mode4.c b/pico/mode4.c
index 0525b935..9d9235ab 100644
--- a/pico/mode4.c
+++ b/pico/mode4.c
@@ -465,6 +465,7 @@ static void DrawDisplayM2(int scanline)
 /*===============*/
 
 static void FinalizeLineRGB555SMS(int line);
+static void FinalizeLine8bitSMS(int line);
 
 void PicoFrameStartSMS(void)
 {
@@ -513,6 +514,12 @@ void PicoFrameStartSMS(void)
 
   Pico.est.HighCol = HighColBase + screen_offset * HighColIncrement;
   Pico.est.DrawLineDest = (char *)DrawLineDestBase + screen_offset * DrawLineDestIncrement;
+
+  if (FinalizeLineSMS == FinalizeLine8bitSMS) {
+    Pico.est.SonicPalCount = 0;
+    Pico.m.dirtyPal = (Pico.m.dirtyPal ? 2 : 0);
+    memcpy(Pico.est.SonicPal, PicoMem.cram, 0x20*2);
+  }
 }
 
 void PicoLineSMS(int line)
@@ -557,32 +564,45 @@ static u16 tmspal[32] = {
 
 void PicoDoHighPal555SMS(void)
 {
-  u32 *spal=(void *)PicoMem.cram;
-  u32 *dpal=(void *)Pico.est.HighPal;
+  u32 *spal = (void *)Pico.est.SonicPal;
+  u32 *dpal = (void *)Pico.est.HighPal;
+  unsigned int cnt = Pico.est.SonicPalCount+1;
   unsigned int t;
-  int i;
+  int i, j;
+ 
+  if (FinalizeLineSMS != FinalizeLine8bitSMS || Pico.m.dirtyPal == 2)
+    Pico.m.dirtyPal = 0;
 
-  Pico.m.dirtyPal = 0;
-  if (!(Pico.video.reg[0] & 0x4))
+  // use hardware palette for 16bit accurate mode
+  if (FinalizeLineSMS == FinalizeLineRGB555SMS)
+    spal = (void *)PicoMem.cram;
+
+  // fixed palette in TMS modes
+  if (!(Pico.video.reg[0] & 0x4)) {
     spal = (u32 *)tmspal;
+    cnt = 1;
+  }
 
   /* SMS 6 bit cram data was already converted to MD/GG format by vdp write,
    * hence GG/SMS/TMS can all be handled the same here */
-  for (i = 0x20/2; i > 0; i--, spal++, dpal++) { 
-    t = *spal;
+  for (j = cnt; j > 0; j--) {
+    for (i = 0x20/2; i > 0; i--, spal++, dpal++) { 
+     t = *spal;
 #if defined(USE_BGR555)
-    t = ((t & 0x000f000f)<< 1) | ((t & 0x00f000f0)<<2) | ((t & 0x0f000f00)<<3);
-    t |= (t >> 4) & 0x04210421;
+     t = ((t & 0x000f000f)<< 1) | ((t & 0x00f000f0)<<2) | ((t & 0x0f000f00)<<3);
+     t |= (t >> 4) & 0x04210421;
 #elif defined(USE_BGR565)
-    t = ((t & 0x000f000f)<< 1) | ((t & 0x00f000f0)<<3) | ((t & 0x0f000f00)<<4);
-    t |= (t >> 4) & 0x08610861;
+     t = ((t & 0x000f000f)<< 1) | ((t & 0x00f000f0)<<3) | ((t & 0x0f000f00)<<4);
+     t |= (t >> 4) & 0x08610861;
 #else
-    t = ((t & 0x000f000f)<<12) | ((t & 0x00f000f0)<<3) | ((t & 0x0f000f00)>>7);
-    t |= (t >> 4) & 0x08610861;
+     t = ((t & 0x000f000f)<<12) | ((t & 0x00f000f0)<<3) | ((t & 0x0f000f00)>>7);
+     t |= (t >> 4) & 0x08610861;
 #endif
-    *dpal = t;
+     *dpal = t;
+    }
+    memcpy(dpal, dpal-0x20/2, 0x20*2); // for prio bit
+    spal += 0x20/2, dpal += 0x20/2;
   }
-  memcpy(&Pico.est.HighPal[0x20], Pico.est.HighPal, 0x20*2); // for prio bit
   Pico.est.HighPal[0xe0] = 0;
 }
 
diff --git a/pico/sms.c b/pico/sms.c
index 78fce13b..7940f5fb 100644
--- a/pico/sms.c
+++ b/pico/sms.c
@@ -21,13 +21,15 @@ extern void YM2413_dataWrite(unsigned data);
 
 
 static unsigned short ymflag = 0xffff;
+static u8 vdp_buffer;
 
 static unsigned char vdp_data_read(void)
 {
   struct PicoVideo *pv = &Pico.video;
   unsigned char d;
 
-  d = PicoMem.vramb[MEM_LE2(pv->addr)];
+  d = vdp_buffer;
+  vdp_buffer = PicoMem.vramb[MEM_LE2(pv->addr)];
   pv->addr = (pv->addr + 1) & 0x3fff;
   pv->pending = 0;
   return d;
@@ -54,12 +56,14 @@ static void vdp_data_write(unsigned char d)
   if (pv->type == 3) {
     if (Pico.m.hardware & 0x1) { // GG, same layout as MD
       unsigned a = pv->addr & 0x3f;
-      if (a & 0x1) d &= 0x0f;
-      if (((u8 *)PicoMem.cram)[MEM_LE2(a)] != d) Pico.m.dirtyPal = 1;
-      ((u8 *)PicoMem.cram)[MEM_LE2(a)] = d;
+      if (a & 0x1) { // write complete color on high byte write
+        u16 c = ((d&0x0f) << 8) | vdp_buffer;
+        if (PicoMem.cram[a >> 1] != c) Pico.m.dirtyPal = 1;
+        PicoMem.cram[a >> 1] = c;
+      }
     } else { // SMS, convert to MD layout (00BbGgRr to 0000BbBbGgGgRrRr)
       unsigned a = pv->addr & 0x1f;
-      u16 c = (d&0x30)*0x40 + (d&0x0c)*0x10 + (d&0x03)*0x04;
+      u16 c = ((d&0x30)<<6) + ((d&0x0c)<<4) + ((d&0x03)<<2);
       if (PicoMem.cram[a] != (c | (c>>2))) Pico.m.dirtyPal = 1;
       PicoMem.cram[a] = c | (c>>2);
     }
@@ -68,6 +72,7 @@ static void vdp_data_write(unsigned char d)
   }
   pv->addr = (pv->addr + 1) & 0x3fff;
 
+  vdp_buffer = d;
   pv->pending = 0;
 }
 
@@ -95,14 +100,18 @@ static void vdp_ctl_write(u8 d)
   struct PicoVideo *pv = &Pico.video;
 
   if (pv->pending) {
-    if ((d >> 6) == 2) {
+    pv->type = d >> 6;
+    if (pv->type == 2) {
       elprintf(EL_IO, "  VDP r%02x=%02x", d & 0x0f, pv->addr & 0xff);
       if (pv->reg[d & 0x0f] != (u8)pv->addr)
         vdp_reg_write(pv, d & 0x0f, pv->addr);
     }
-    pv->type = d >> 6;
     pv->addr &= 0x00ff;
     pv->addr |= (d & 0x3f) << 8;
+    if (pv->type == 0) {
+      vdp_buffer = PicoMem.vramb[MEM_LE2(pv->addr)];
+      pv->addr = (pv->addr + 1) & 0x3fff;
+    }
   } else {
     pv->addr &= 0x3f00;
     pv->addr |= d;
@@ -147,8 +156,12 @@ static unsigned char z80_sms_in(unsigned short a)
         elprintf(EL_HVCNT, "V counter read: %02x", d);
         break;
 
-      case 0x41: /* H counter */
-        d = Pico.m.rotate++;
+      case 0x41: /* H counter, TODO: latched by toggle of pad TH line */
+        // 171 slots per scanline of 228 clocks, runs from 0x85-0x93,0xe9-0x84
+#define CYC2SLOT (256 * 171/228) // cycles to slot factor in Q8
+        d = 228-z80_cyclesLeft;
+        if (d <= 19)	d = (( d     * CYC2SLOT)>>8) + 0x85;
+        else		d = (((d-20) * CYC2SLOT)>>8) + 0xe9;
         elprintf(EL_HVCNT, "H counter read: %02x", d);
         break;
 
@@ -404,7 +417,7 @@ void PicoFrameMS(void)
   struct PicoVideo *pv = &Pico.video;
   int is_pal = Pico.m.pal;
   int lines = is_pal ? 313 : 262;
-  int cycles_line = is_pal ? 58020 : 58293; /* (226.6 : 227.7) * 256 */
+  int cycles_line = 228;
   int cycles_done = 0, cycles_aim = 0;
   int skip = PicoIn.skipFrame;
   int lines_vis = 192;
@@ -454,8 +467,8 @@ void PicoFrameMS(void)
     }
 
     cycles_aim += cycles_line;
-    Pico.t.z80c_aim = cycles_aim >> 8;
-    cycles_done += z80_run((cycles_aim - cycles_done) >> 8) << 8;
+    Pico.t.z80c_aim = cycles_aim;
+    cycles_done += z80_run(cycles_aim - cycles_done);
   }
 
   if (PicoIn.sndOut)