diff --git a/video/decode/vd_lavc.c b/video/decode/vd_lavc.c
index e07eecc2d2..6c356e2e01 100644
--- a/video/decode/vd_lavc.c
+++ b/video/decode/vd_lavc.c
@@ -860,7 +860,7 @@ static int get_buffer2_direct(AVCodecContext *avctx, AVFrame *pic, int flags)
 
     // We assume that different alignments are just different power-of-2s.
     // Thus, a higher alignment always satisfies a lower alignment.
-    int stride_align = 0;
+    int stride_align = MP_IMAGE_BYTE_ALIGN;
     for (int n = 0; n < AV_NUM_DATA_POINTERS; n++)
         stride_align = MPMAX(stride_align, linesize_align[n]);
 
diff --git a/video/img_format.c b/video/img_format.c
index f7790aa290..be81e15b41 100644
--- a/video/img_format.c
+++ b/video/img_format.c
@@ -539,7 +539,7 @@ int main(int argc, char **argv)
             fr->format = fmt;
             fr->width = 128;
             fr->height = 128;
-            int err = av_frame_get_buffer(fr, SWS_MIN_BYTE_ALIGN);
+            int err = av_frame_get_buffer(fr, MP_IMAGE_BYTE_ALIGN);
             assert(err >= 0);
             struct mp_image *mpi = mp_image_alloc(mpfmt, fr->width, fr->height);
             assert(mpi);
diff --git a/video/mp_image.c b/video/mp_image.c
index f846b0d3d3..74ca80db21 100644
--- a/video/mp_image.c
+++ b/video/mp_image.c
@@ -174,7 +174,7 @@ static bool mp_image_alloc_planes(struct mp_image *mpi)
     assert(!mpi->planes[0]);
     assert(!mpi->bufs[0]);
 
-    int align = SWS_MIN_BYTE_ALIGN;
+    int align = MP_IMAGE_BYTE_ALIGN;
 
     int size = mp_image_get_alloc_size(mpi->imgfmt, mpi->w, mpi->h, align);
     if (size < 0)
diff --git a/video/mp_image.h b/video/mp_image.h
index 4727ed6f8d..0eed3eb43c 100644
--- a/video/mp_image.h
+++ b/video/mp_image.h
@@ -28,6 +28,12 @@
 #include "csputils.h"
 #include "video/img_format.h"
 
+// Assumed minimum align needed for image allocation. It's notable that FFmpeg's
+// libraries except libavcodec don't really know what alignment they want.
+// Things will randomly crash or get slower if the alignment is not satisfied.
+// Whatever. This value should be pretty safe with current CPU architectures.
+#define MP_IMAGE_BYTE_ALIGN 64
+
 #define MP_IMGFIELD_TOP_FIRST 0x02
 #define MP_IMGFIELD_REPEAT_FIRST 0x04
 #define MP_IMGFIELD_INTERLACED 0x20
diff --git a/video/sws_utils.h b/video/sws_utils.h
index 41472b5ecf..629e6be2dd 100644
--- a/video/sws_utils.h
+++ b/video/sws_utils.h
@@ -11,7 +11,7 @@ struct mpv_global;
 // libswscale currently requires 16 bytes alignment for row pointers and
 // strides. Otherwise, it will print warnings and use slow codepaths.
 // Guaranteed to be a power of 2 and > 1.
-#define SWS_MIN_BYTE_ALIGN 16
+#define SWS_MIN_BYTE_ALIGN MP_IMAGE_BYTE_ALIGN
 
 extern const int mp_sws_hq_flags;
 extern const int mp_sws_fast_flags;