diff --git a/Core/HW/BufferQueue.h b/Core/HW/BufferQueue.h
index 79e7072f198e..f4231062f3c3 100644
--- a/Core/HW/BufferQueue.h
+++ b/Core/HW/BufferQueue.h
@@ -124,6 +124,22 @@ struct BufferQueue {
 		return bytesgot;
 	}
 
+	bool containsAuAfter(int off) {
+		int avail = getQueueSize();
+		for (int i = off; i <= avail - 3; ++i) {
+			if (peek(i) != 0) {
+				continue;
+			}
+
+			if (peek(i + 1) == 1 && peek(i + 2) == 9) {
+				NOTICE_LOG(HLE, "found AU at offset %d: %d", i, i > 0 ? peek(i - 1) : 1337);
+				return true;
+			}
+		}
+
+		return false;
+	}
+
 	void DoState(PointerWrap &p) {
 		auto s = p.Section("BufferQueue", 0, 1);
 
@@ -172,6 +188,11 @@ struct BufferQueue {
 		return pts;
 	}
 
+	unsigned char peek(int i) {
+		int off = start + i % bufQueueSize;
+		return bufQueue[i];
+	}
+
 	unsigned char* bufQueue;
 	int start, end;
 	int bufQueueSize;
diff --git a/Core/HW/MediaEngine.cpp b/Core/HW/MediaEngine.cpp
index b89232971b6f..9e501ed1c692 100644
--- a/Core/HW/MediaEngine.cpp
+++ b/Core/HW/MediaEngine.cpp
@@ -553,11 +553,40 @@ void MediaEngine::updateSwsFormat(int videoPixelMode) {
 #endif
 }
 
+void MediaEngine::HandleDecodedFrame(AVCodecContext *ctx, int videoPixelMode, bool skipFrame) {
+	if (!m_pFrameRGB) {
+		setVideoDim();
+	}
+
+	if (m_pFrameRGB && !skipFrame) {
+		updateSwsFormat(videoPixelMode);
+		// TODO: Technically we could set this to frameWidth instead of m_desWidth for better perf.
+		// Update the linesize for the new format too.  We started with the largest size, so it should fit.
+		m_pFrameRGB->linesize[0] = getPixelFormatBytes(videoPixelMode) * m_desWidth;
+
+		sws_scale(m_sws_ctx, m_pFrame->data, m_pFrame->linesize, 0,
+			ctx->height, m_pFrameRGB->data, m_pFrameRGB->linesize);
+	}
+
+	if (av_frame_get_best_effort_timestamp(m_pFrame) != AV_NOPTS_VALUE)
+		m_videopts = av_frame_get_best_effort_timestamp(m_pFrame) + av_frame_get_pkt_duration(m_pFrame) - m_firstTimeStamp;
+	else
+		m_videopts += av_frame_get_pkt_duration(m_pFrame);
+}
+
 bool MediaEngine::stepVideo(int videoPixelMode, bool skipFrame) {
 #ifdef USE_FFMPEG
 	auto codecIter = m_pCodecCtxs.find(m_videoStream);
 	AVCodecContext *m_pCodecCtx = codecIter == m_pCodecCtxs.end() ? 0 : codecIter->second;
 
+	if (m_desWidth == 0 || m_desHeight == 0) {
+		// Sometimes there is no separator, let's assume 20kb is always enough.
+		// Less than 20kb is not enough for Valkyrie Profile.
+		if (!m_pdata->containsAuAfter(m_mpegheaderSize) && m_pdata->getQueueSize() < 20480) {
+			return false;
+		}
+	}
+
 	if (!m_pFormatCtx)
 		return false;
 	if (!m_pCodecCtx)
@@ -580,23 +609,7 @@ bool MediaEngine::stepVideo(int videoPixelMode, bool skipFrame) {
 
 			int result = avcodec_decode_video2(m_pCodecCtx, m_pFrame, &frameFinished, &packet);
 			if (frameFinished) {
-				if (!m_pFrameRGB) {
-					setVideoDim();
-				}
-				if (m_pFrameRGB && !skipFrame) {
-					updateSwsFormat(videoPixelMode);
-					// TODO: Technically we could set this to frameWidth instead of m_desWidth for better perf.
-					// Update the linesize for the new format too.  We started with the largest size, so it should fit.
-					m_pFrameRGB->linesize[0] = getPixelFormatBytes(videoPixelMode) * m_desWidth;
-
-					sws_scale(m_sws_ctx, m_pFrame->data, m_pFrame->linesize, 0,
-						m_pCodecCtx->height, m_pFrameRGB->data, m_pFrameRGB->linesize);
-				}
-
-				if (av_frame_get_best_effort_timestamp(m_pFrame) != AV_NOPTS_VALUE)
-					m_videopts = av_frame_get_best_effort_timestamp(m_pFrame) + av_frame_get_pkt_duration(m_pFrame) - m_firstTimeStamp;
-				else
-					m_videopts += av_frame_get_pkt_duration(m_pFrame);
+				HandleDecodedFrame(m_pCodecCtx, videoPixelMode, skipFrame);
 				bGetFrame = true;
 			}
 			if (result <= 0 && dataEnd) {
diff --git a/Core/HW/MediaEngine.h b/Core/HW/MediaEngine.h
index 6561176b03af..a59dd82cf930 100644
--- a/Core/HW/MediaEngine.h
+++ b/Core/HW/MediaEngine.h
@@ -98,6 +98,10 @@ class MediaEngine
 	void updateSwsFormat(int videoPixelMode);
 	int getNextAudioFrame(u8 **buf, int *headerCode1, int *headerCode2);
 
+#ifdef USE_FFMPEG
+	void HandleDecodedFrame(AVCodecContext *ctx, int videoPixelMode, bool skipFrame);
+#endif
+
 public:  // TODO: Very little of this below should be public.
 
 	// Video ffmpeg context - not used for audio