diff --git a/filters/f_swscale.c b/filters/f_swscale.c
index 89fc2bb3fc..8e684d5507 100644
--- a/filters/f_swscale.c
+++ b/filters/f_swscale.c
@@ -105,6 +105,7 @@ static void sws_process(struct mp_filter *f)
         goto error;
 
     mp_image_copy_attributes(dst, src);
+    mp_image_setfmt(dst, dstfmt);
 
     if (s->use_out_params)
         dst->params = s->out_params;
diff --git a/test/img_format.c b/test/img_format.c
index 835c34acaf..c0a6480cac 100644
--- a/test/img_format.c
+++ b/test/img_format.c
@@ -84,6 +84,24 @@ int main(int argc, char *argv[])
                 fprintf(f, " endian_bytes=%d", 1 << d.endian_shift);
             for (int x = 0; x < MP_NUM_COMPONENTS; x++) {
                 struct mp_imgfmt_comp_desc cm = d.comps[x];
+                if (avd && x < avd->nb_components) {
+                    const AVComponentDescriptor *cd = &avd->comp[x];
+                    // Validate only if padding is not 0. fill_pixdesc_layout()
+                    // uses some heuristics to convert AVComponentDescriptor
+                    // to mp_imgfmt_comp_desc. Some formats, mostly packed ones,
+                    // are not fully supported. See fill_pixdesc_layout()
+                    // for more details.
+                    if (cm.pad) {
+                        struct pl_bit_encoding be = {
+                            .sample_depth = cm.size,
+                            .color_depth = cm.size - abs(cm.pad),
+                            .bit_shift = MPMAX(0, cm.pad),
+                        };
+                        mp_require(be.color_depth == cd->depth);
+                        mp_require(be.sample_depth == MP_ALIGN_UP(cd->depth, 8));
+                        mp_require(be.bit_shift == cd->shift);
+                    }
+                }
                 fprintf(f, " {");
                 if (cm.plane == n) {
                     if (cm.size) {
diff --git a/video/mp_image.c b/video/mp_image.c
index 7890b9f9f4..1c629a6287 100644
--- a/video/mp_image.c
+++ b/video/mp_image.c
@@ -194,6 +194,17 @@ void mp_image_setfmt(struct mp_image *mpi, int out_fmt)
     mpi->fmt = fmt;
     mpi->imgfmt = fmt.id;
     mpi->num_planes = fmt.num_planes;
+    mpi->params.repr.alpha = (fmt.flags & MP_IMGFLAG_ALPHA) ? PL_ALPHA_INDEPENDENT
+#if PL_API_VER >= 344
+                                                            : PL_ALPHA_NONE;
+#else
+                                                            : PL_ALPHA_UNKNOWN;
+#endif
+    mpi->params.repr.bits = (struct pl_bit_encoding) {
+        .sample_depth = fmt.comps[0].size,
+        .color_depth = fmt.comps[0].size - abs(fmt.comps[0].pad),
+        .bit_shift = MPMAX(0, fmt.comps[0].pad),
+    };
 }
 
 static void mp_image_destructor(void *ptr)
@@ -1073,10 +1084,8 @@ struct mp_image *mp_image_from_av_frame(struct AVFrame *src)
     if (src->repeat_pict == 1)
         dst->fields |= MP_IMGFIELD_REPEAT_FIRST;
 
-    dst->params.repr = (struct pl_color_repr){
-        .sys = pl_system_from_av(src->colorspace),
-        .levels = pl_levels_from_av(src->color_range),
-    };
+    dst->params.repr.sys = pl_system_from_av(src->colorspace);
+    dst->params.repr.levels = pl_levels_from_av(src->color_range);
 
     dst->params.color = (struct pl_color_space){
         .primaries = pl_primaries_from_av(src->color_primaries),