diff mbox series

[FFmpeg-devel,v5,4/7] libavcodec/webp: add support for animated WebP decoding

Message ID 20231120192211.77775-5-thilo.borgmann@mail.de
State New
Headers show
Series webp: add support for animated WebP decoding | expand

Checks

Context Check Description
yinshiyou/make_loongarch64 fail Make failed
andriy/make_x86 fail Make failed

Commit Message

Thilo Borgmann Nov. 20, 2023, 7:22 p.m. UTC
From: Josef Zlomek <josef@pex.com>

Fixes: 4907

Adds support for decoding of animated WebP.

The WebP decoder adds the animation related features according to the specs:
https://developers.google.com/speed/webp/docs/riff_container#animation
The frames of the animation may be smaller than the image canvas.
Therefore, the frame is decoded to a temporary frame,
then it is blended into the canvas, the canvas is copied to the output frame,
and finally the frame is disposed from the canvas.

The output to AV_PIX_FMT_YUVA420P/AV_PIX_FMT_YUV420P is still supported.
The background color is specified only as BGRA in the WebP file
so it is converted to YUVA if YUV formats are output.

Signed-off-by: Josef Zlomek <josef@pex.com>
---
 Changelog               |   1 +
 libavcodec/codec_desc.c |   3 +-
 libavcodec/version.h    |   2 +-
 libavcodec/webp.c       | 763 ++++++++++++++++++++++++++++++++++++----
 4 files changed, 700 insertions(+), 69 deletions(-)

Comments

James Almer Nov. 21, 2023, 1:14 a.m. UTC | #1
On 11/20/2023 4:22 PM, Thilo Borgmann via ffmpeg-devel wrote:
> +    if (*got_frame) {
> +        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
> +            // no animation, output the decoded frame
> +            av_frame_move_ref(p, s->frame);
> +            ret = ff_attach_decode_data(p);
> +            if (ret < 0)
> +                return ret;

If this frame's buffers were allocated with ff_get_buffer() (or its 
threaded version), then a call to ff_attach_decode_data() is not necessary.
Cosmin Stejerean Nov. 22, 2023, 2:54 a.m. UTC | #2
On Nov 20, 2023, at 5:14 PM, James Almer <jamrial@gmail.com> wrote:

On 11/20/2023 4:22 PM, Thilo Borgmann via ffmpeg-devel wrote:
+    if (*got_frame) {
+        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+            // no animation, output the decoded frame
+            av_frame_move_ref(p, s->frame);
+            ret = ff_attach_decode_data(p);
+            if (ret < 0)
+                return ret;

If this frame's buffers were allocated with ff_get_buffer() (or its threaded version), then a call to ff_attach_decode_data() is not necessary.

The frame returned by avcodec_receive_frame doesn’t have private_ref set on it, and eventually this null private_ref would get propagated to the other frame objects (even if allocated with ff_get_buffer) by av_frame_copy_props, and later on this would fail the assert that frames returned by codecs with AV_CODEC_CAP_DR1 have private_ref set on them.

However this patch is dealing with it in the wrong place, moving ff_attach_decode_data to right after avcodec_receive_frame would both make this more clear and fix the assert issues (currently there’s an unnecessary call to ff_attach_decode_data on an already allocated frame in a code path that doesn’t go through vp8_lossy_decode_frame).

I verified that with that change on top of the current patch set it would run the webp fate tests cleanly. I’m not sure if this is the best way to properly reference a frame that came from avcodec_receive_frame inside of another codec. But it does seem to work and a cursory search didn’t reveal a better API for this.


- Cosmin
Thilo Borgmann Nov. 22, 2023, 7:36 p.m. UTC | #3
Am 22.11.23 um 03:54 schrieb Cosmin Stejerean via ffmpeg-devel:
> On Nov 20, 2023, at 5:14 PM, James Almer <jamrial@gmail.com> wrote:
> 
> On 11/20/2023 4:22 PM, Thilo Borgmann via ffmpeg-devel wrote:
> +    if (*got_frame) {
> +        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
> +            // no animation, output the decoded frame
> +            av_frame_move_ref(p, s->frame);
> +            ret = ff_attach_decode_data(p);
> +            if (ret < 0)
> +                return ret;
> 
> If this frame's buffers were allocated with ff_get_buffer() (or its threaded version), then a call to ff_attach_decode_data() is not necessary.
> 
> The frame returned by avcodec_receive_frame doesn’t have private_ref set on it, and eventually this null private_ref would get propagated to the other frame objects (even if allocated with ff_get_buffer) by av_frame_copy_props, and later on this would fail the assert that frames returned by codecs with AV_CODEC_CAP_DR1 have private_ref set on them.


> However this patch is dealing with it in the wrong place, moving ff_attach_decode_data to right after avcodec_receive_frame would both make this more clear and fix the assert issues (currently there’s an unnecessary call to ff_attach_decode_data on an already allocated frame in a code path that doesn’t go through vp8_lossy_decode_frame).
> 
> I verified that with that change on top of the current patch set it would run the webp fate tests cleanly. I’m not sure if this is the best way to properly reference a frame that came from avcodec_receive_frame inside of another codec. But it does seem to work and a cursory search didn’t reveal a better API for this.

Moving ff_attach_decode_data() right after avcodec_receive_frame() solved FATE & the asserts even for --assert-level=2.

Will post a v6 according to this soon, unless there's more feedback on the buffer creation.

Thanks,
Thilo
diff mbox series

Patch

diff --git a/Changelog b/Changelog
index 7d79be1c3e..8f34660767 100644
--- a/Changelog
+++ b/Changelog
@@ -43,6 +43,7 @@  version 6.1:
 - ffprobe XML output schema changed to account for multiple
   variable-fields elements within the same parent element
 - ffprobe -output_format option added as an alias of -of
+- animated WebP parser/decoder
 
 
 version 6.0:
diff --git a/libavcodec/codec_desc.c b/libavcodec/codec_desc.c
index 432a9c9ea6..2ab975f8d7 100644
--- a/libavcodec/codec_desc.c
+++ b/libavcodec/codec_desc.c
@@ -1259,8 +1259,7 @@  static const AVCodecDescriptor codec_descriptors[] = {
         .type      = AVMEDIA_TYPE_VIDEO,
         .name      = "webp",
         .long_name = NULL_IF_CONFIG_SMALL("WebP"),
-        .props     = AV_CODEC_PROP_INTRA_ONLY | AV_CODEC_PROP_LOSSY |
-                     AV_CODEC_PROP_LOSSLESS,
+        .props     = AV_CODEC_PROP_LOSSY | AV_CODEC_PROP_LOSSLESS,
         .mime_types= MT("image/webp"),
     },
     {
diff --git a/libavcodec/version.h b/libavcodec/version.h
index 0ef6c991f3..0a91c6b916 100644
--- a/libavcodec/version.h
+++ b/libavcodec/version.h
@@ -30,7 +30,7 @@ 
 #include "version_major.h"
 
 #define LIBAVCODEC_VERSION_MINOR  34
-#define LIBAVCODEC_VERSION_MICRO 100
+#define LIBAVCODEC_VERSION_MICRO 101
 
 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
                                                LIBAVCODEC_VERSION_MINOR, \
diff --git a/libavcodec/webp.c b/libavcodec/webp.c
index 286e7c8b73..52ecdb45c5 100644
--- a/libavcodec/webp.c
+++ b/libavcodec/webp.c
@@ -35,12 +35,16 @@ 
  * Exif metadata
  * ICC profile
  *
+ * @author Josef Zlomek, Pexeso Inc. <josef@pex.com>
+ * Animation
+ *
  * Unimplemented:
- *   - Animation
  *   - XMP metadata
  */
 
+#include "libavcodec/packet.h"
 #include "libavutil/imgutils.h"
+#include "libavutil/colorspace.h"
 
 #define BITSTREAM_READER_LE
 #include "avcodec.h"
@@ -192,9 +196,12 @@  typedef struct ImageContext {
 typedef struct WebPContext {
     VP8Context v;                       /* VP8 Context used for lossy decoding */
     GetBitContext gb;                   /* bitstream reader for main image chunk */
+    ThreadFrame canvas_frame;           /* ThreadFrame for canvas */
+    AVFrame *frame;                     /* AVFrame for decoded frame */
     AVFrame *alpha_frame;               /* AVFrame for alpha data decompressed from VP8L */
     AVPacket *pkt;                      /* AVPacket to be passed to the underlying VP8 decoder */
     AVCodecContext *avctx;              /* parent AVCodecContext */
+    AVCodecContext *avctx_vp8;          /* wrapper context for VP8 decoder */
     int initialized;                    /* set once the VP8 context is initialized */
     int has_alpha;                      /* has a separate alpha chunk */
     enum AlphaCompression alpha_compression; /* compression type for alpha chunk */
@@ -203,9 +210,24 @@  typedef struct WebPContext {
     int alpha_data_size;                /* alpha chunk data size */
     int has_exif;                       /* set after an EXIF chunk has been processed */
     int has_iccp;                       /* set after an ICCP chunk has been processed */
-    int width;                          /* image width */
-    int height;                         /* image height */
-    int lossless;                       /* indicates lossless or lossy */
+    int vp8x_flags;                     /* global flags from VP8X chunk */
+    int canvas_width;                   /* canvas width */
+    int canvas_height;                  /* canvas height */
+    int anmf_flags;                     /* frame flags from ANMF chunk */
+    int width;                          /* frame width */
+    int height;                         /* frame height */
+    int pos_x;                          /* frame position X */
+    int pos_y;                          /* frame position Y */
+    int prev_anmf_flags;                /* previous frame flags from ANMF chunk */
+    int prev_width;                     /* previous frame width */
+    int prev_height;                    /* previous frame height */
+    int prev_pos_x;                     /* previous frame position X */
+    int prev_pos_y;                     /* previous frame position Y */
+    int await_progress;                 /* value of progress to wait for */
+    uint8_t background_argb[4];         /* background color in ARGB format */
+    uint8_t background_yuva[4];         /* background color in YUVA format */
+    const uint8_t *background_data[4];  /* "planes" for background color in YUVA format */
+    uint8_t transparent_yuva[4];        /* transparent black in YUVA format */
 
     int nb_transforms;                  /* number of transforms */
     enum TransformType transforms[4];   /* transformations used in the image, in order */
@@ -569,7 +591,7 @@  static int decode_entropy_coded_image(WebPContext *s, enum ImageRole role,
     img->frame->height = h;
 
     if (role == IMAGE_ROLE_ARGB && !img->is_alpha_primary) {
-        ret = ff_thread_get_buffer(s->avctx, img->frame, 0);
+        ret = ff_get_buffer(s->avctx, img->frame, 0);
     } else
         ret = av_frame_get_buffer(img->frame, 1);
     if (ret < 0)
@@ -1067,7 +1089,7 @@  static int apply_color_indexing_transform(WebPContext *s)
     return 0;
 }
 
-static void update_canvas_size(AVCodecContext *avctx, int w, int h)
+static void update_frame_size(AVCodecContext *avctx, int w, int h)
 {
     WebPContext *s = avctx->priv_data;
     if (s->width && s->width != w) {
@@ -1090,7 +1112,6 @@  static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
     int w, h, ret, i, used;
 
     if (!is_alpha_chunk) {
-        s->lossless = 1;
         avctx->pix_fmt = AV_PIX_FMT_ARGB;
     }
 
@@ -1107,7 +1128,7 @@  static int vp8_lossless_decode_frame(AVCodecContext *avctx, AVFrame *p,
         w = get_bits(&s->gb, 14) + 1;
         h = get_bits(&s->gb, 14) + 1;
 
-        update_canvas_size(avctx, w, h);
+        update_frame_size(avctx, w, h);
 
         ret = ff_set_dimensions(avctx, s->width, s->height);
         if (ret < 0)
@@ -1299,12 +1320,12 @@  static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
     int ret;
 
     if (!s->initialized) {
-        ff_vp8_decode_init(avctx);
+        VP8Context *s_vp8 = s->avctx_vp8->priv_data;
+        s_vp8->actually_webp = 1;
         s->initialized = 1;
-        s->v.actually_webp = 1;
     }
     avctx->pix_fmt = s->has_alpha ? AV_PIX_FMT_YUVA420P : AV_PIX_FMT_YUV420P;
-    s->lossless = 0;
+    s->avctx_vp8->pix_fmt = s->has_alpha ? AV_PIX_FMT_YUVA420P : AV_PIX_FMT_YUV420P;
 
     if (data_size > INT_MAX) {
         av_log(avctx, AV_LOG_ERROR, "unsupported chunk size\n");
@@ -1315,14 +1336,20 @@  static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
     s->pkt->data = data_start;
     s->pkt->size = data_size;
 
-    ret = ff_vp8_decode_frame(avctx, p, got_frame, s->pkt);
-    if (ret < 0)
+    ret = avcodec_send_packet(s->avctx_vp8, s->pkt);
+    if (ret < 0) {
+        av_log(avctx, AV_LOG_ERROR, "Error submitting a packet for decoding\n");
         return ret;
+    }
 
-    if (!*got_frame)
+    ret = avcodec_receive_frame(s->avctx_vp8, p);
+    if (ret < 0) {
+        av_log(avctx, AV_LOG_ERROR, "VP8 decoding error: %s.\n", av_err2str(ret));
         return AVERROR_INVALIDDATA;
+    }
 
-    update_canvas_size(avctx, avctx->width, avctx->height);
+    *got_frame = 1;
+    update_frame_size(avctx, s->avctx_vp8->width, s->avctx_vp8->height);
 
     if (s->has_alpha) {
         ret = vp8_lossy_decode_alpha(avctx, p, s->alpha_data,
@@ -1333,40 +1360,17 @@  static int vp8_lossy_decode_frame(AVCodecContext *avctx, AVFrame *p,
     return ret;
 }
 
-static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
-                             int *got_frame, AVPacket *avpkt)
+int init_canvas_frame(WebPContext *s, int format, int key_frame);
+
+static int webp_decode_frame_common(AVCodecContext *avctx, uint8_t *data, int size,
+                                    int *got_frame, int key_frame)
 {
     WebPContext *s = avctx->priv_data;
     GetByteContext gb;
     int ret;
     uint32_t chunk_type, chunk_size;
-    int vp8x_flags = 0;
-
-    s->avctx     = avctx;
-    s->width     = 0;
-    s->height    = 0;
-    *got_frame   = 0;
-    s->has_alpha = 0;
-    s->has_exif  = 0;
-    s->has_iccp  = 0;
-    bytestream2_init(&gb, avpkt->data, avpkt->size);
-
-    if (bytestream2_get_bytes_left(&gb) < 12)
-        return AVERROR_INVALIDDATA;
 
-    if (bytestream2_get_le32(&gb) != MKTAG('R', 'I', 'F', 'F')) {
-        av_log(avctx, AV_LOG_ERROR, "missing RIFF tag\n");
-        return AVERROR_INVALIDDATA;
-    }
-
-    chunk_size = bytestream2_get_le32(&gb);
-    if (bytestream2_get_bytes_left(&gb) < chunk_size)
-        return AVERROR_INVALIDDATA;
-
-    if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
-        av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
-        return AVERROR_INVALIDDATA;
-    }
+    bytestream2_init(&gb, data, size);
 
     while (bytestream2_get_bytes_left(&gb) > 8) {
         char chunk_str[5] = { 0 };
@@ -1377,6 +1381,10 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             return AVERROR_INVALIDDATA;
         chunk_size += chunk_size & 1;
 
+        // we need to dive into RIFF chunk
+        if (chunk_type == MKTAG('R', 'I', 'F', 'F'))
+            chunk_size = 4;
+
         if (bytestream2_get_bytes_left(&gb) < chunk_size) {
            /* we seem to be running out of data, but it could also be that the
               bitstream has trailing junk leading to bogus chunk_size. */
@@ -1384,10 +1392,26 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
         }
 
         switch (chunk_type) {
+        case MKTAG('R', 'I', 'F', 'F'):
+            if (bytestream2_get_le32(&gb) != MKTAG('W', 'E', 'B', 'P')) {
+                av_log(avctx, AV_LOG_ERROR, "missing WEBP tag\n");
+                return AVERROR_INVALIDDATA;
+            }
+            s->vp8x_flags    = 0;
+            s->canvas_width  = 0;
+            s->canvas_height = 0;
+            s->has_exif      = 0;
+            s->has_iccp      = 0;
+            ff_thread_release_ext_buffer(&s->canvas_frame);
+            break;
         case MKTAG('V', 'P', '8', ' '):
             if (!*got_frame) {
-                ret = vp8_lossy_decode_frame(avctx, p, got_frame,
-                                             avpkt->data + bytestream2_tell(&gb),
+                ret = init_canvas_frame(s, AV_PIX_FMT_YUVA420P, key_frame);
+                if (ret < 0)
+                    return ret;
+
+                ret = vp8_lossy_decode_frame(avctx, s->frame, got_frame,
+                                             data + bytestream2_tell(&gb),
                                              chunk_size);
                 if (ret < 0)
                     return ret;
@@ -1396,8 +1420,12 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             break;
         case MKTAG('V', 'P', '8', 'L'):
             if (!*got_frame) {
-                ret = vp8_lossless_decode_frame(avctx, p, got_frame,
-                                                avpkt->data + bytestream2_tell(&gb),
+                ret = init_canvas_frame(s, AV_PIX_FMT_ARGB, key_frame);
+                if (ret < 0)
+                    return ret;
+
+                ret = vp8_lossless_decode_frame(avctx, s->frame, got_frame,
+                                                data + bytestream2_tell(&gb),
                                                 chunk_size, 0);
                 if (ret < 0)
                     return ret;
@@ -1406,14 +1434,16 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
             bytestream2_skip(&gb, chunk_size);
             break;
         case MKTAG('V', 'P', '8', 'X'):
-            if (s->width || s->height || *got_frame) {
+            if (s->canvas_width || s->canvas_height || *got_frame) {
                 av_log(avctx, AV_LOG_ERROR, "Canvas dimensions are already set\n");
                 return AVERROR_INVALIDDATA;
             }
-            vp8x_flags = bytestream2_get_byte(&gb);
+            s->vp8x_flags = bytestream2_get_byte(&gb);
             bytestream2_skip(&gb, 3);
             s->width  = bytestream2_get_le24(&gb) + 1;
             s->height = bytestream2_get_le24(&gb) + 1;
+            s->canvas_width  = s->width;
+            s->canvas_height = s->height;
             ret = av_image_check_size(s->width, s->height, 0, avctx);
             if (ret < 0)
                 return ret;
@@ -1421,7 +1451,7 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
         case MKTAG('A', 'L', 'P', 'H'): {
             int alpha_header, filter_m, compression;
 
-            if (!(vp8x_flags & VP8X_FLAG_ALPHA)) {
+            if (!(s->vp8x_flags & VP8X_FLAG_ALPHA)) {
                 av_log(avctx, AV_LOG_WARNING,
                        "ALPHA chunk present, but alpha bit not set in the "
                        "VP8X header\n");
@@ -1430,8 +1460,9 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 av_log(avctx, AV_LOG_ERROR, "invalid ALPHA chunk size\n");
                 return AVERROR_INVALIDDATA;
             }
+
             alpha_header       = bytestream2_get_byte(&gb);
-            s->alpha_data      = avpkt->data + bytestream2_tell(&gb);
+            s->alpha_data      = data + bytestream2_tell(&gb);
             s->alpha_data_size = chunk_size - 1;
             bytestream2_skip(&gb, s->alpha_data_size);
 
@@ -1458,14 +1489,13 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 av_log(avctx, AV_LOG_VERBOSE, "Ignoring extra EXIF chunk\n");
                 goto exif_end;
             }
-            if (!(vp8x_flags & VP8X_FLAG_EXIF_METADATA))
+            if (!(s->vp8x_flags & VP8X_FLAG_EXIF_METADATA))
                 av_log(avctx, AV_LOG_WARNING,
                        "EXIF chunk present, but Exif bit not set in the "
                        "VP8X header\n");
 
             s->has_exif = 1;
-            bytestream2_init(&exif_gb, avpkt->data + exif_offset,
-                             avpkt->size - exif_offset);
+            bytestream2_init(&exif_gb, data + exif_offset, size - exif_offset);
             if (ff_tdecode_header(&exif_gb, &le, &ifd_offset) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "invalid TIFF header "
                        "in Exif data\n");
@@ -1478,7 +1508,7 @@  static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
                 goto exif_end;
             }
 
-            av_dict_copy(&p->metadata, exif_metadata, 0);
+            av_dict_copy(&s->frame->metadata, exif_metadata, 0);
 
 exif_end:
             av_dict_free(&exif_metadata);
@@ -1493,21 +1523,64 @@  exif_end:
                 bytestream2_skip(&gb, chunk_size);
                 break;
             }
-            if (!(vp8x_flags & VP8X_FLAG_ICC))
+            if (!(s->vp8x_flags & VP8X_FLAG_ICC))
                 av_log(avctx, AV_LOG_WARNING,
                        "ICCP chunk present, but ICC Profile bit not set in the "
                        "VP8X header\n");
 
             s->has_iccp = 1;
-            sd = av_frame_new_side_data(p, AV_FRAME_DATA_ICC_PROFILE, chunk_size);
+            sd = av_frame_new_side_data(s->frame, AV_FRAME_DATA_ICC_PROFILE, chunk_size);
             if (!sd)
                 return AVERROR(ENOMEM);
 
             bytestream2_get_buffer(&gb, sd->data, chunk_size);
             break;
         }
-        case MKTAG('A', 'N', 'I', 'M'):
+        case MKTAG('A', 'N', 'I', 'M'): {
+            const AVPixFmtDescriptor *desc;
+            int a, r, g, b;
+            if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+                av_log(avctx, AV_LOG_WARNING,
+                       "ANIM chunk present, but animation bit not set in the "
+                       "VP8X header\n");
+            }
+            // background is stored as BGRA, we need ARGB
+            s->background_argb[3] = b = bytestream2_get_byte(&gb);
+            s->background_argb[2] = g = bytestream2_get_byte(&gb);
+            s->background_argb[1] = r = bytestream2_get_byte(&gb);
+            s->background_argb[0] = a = bytestream2_get_byte(&gb);
+
+            // convert the background color to YUVA
+            desc = av_pix_fmt_desc_get(AV_PIX_FMT_YUVA420P);
+            s->background_yuva[desc->comp[0].plane] = RGB_TO_Y_CCIR(r, g, b);
+            s->background_yuva[desc->comp[1].plane] = RGB_TO_U_CCIR(r, g, b, 0);
+            s->background_yuva[desc->comp[2].plane] = RGB_TO_V_CCIR(r, g, b, 0);
+            s->background_yuva[desc->comp[3].plane] = a;
+
+            bytestream2_skip(&gb, 2); // loop count is ignored
+            break;
+        }
         case MKTAG('A', 'N', 'M', 'F'):
+            if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+                av_log(avctx, AV_LOG_WARNING,
+                       "ANMF chunk present, but animation bit not set in the "
+                       "VP8X header\n");
+            }
+            s->pos_x      = bytestream2_get_le24(&gb) * 2;
+            s->pos_y      = bytestream2_get_le24(&gb) * 2;
+            s->width      = bytestream2_get_le24(&gb) + 1;
+            s->height     = bytestream2_get_le24(&gb) + 1;
+            bytestream2_skip(&gb, 3);   // duration
+            s->anmf_flags = bytestream2_get_byte(&gb);
+
+            if (s->width  + s->pos_x > s->canvas_width ||
+                s->height + s->pos_y > s->canvas_height) {
+                av_log(avctx, AV_LOG_ERROR,
+                       "frame does not fit into canvas\n");
+                return AVERROR_INVALIDDATA;
+            }
+            s->vp8x_flags |= VP8X_FLAG_ANIMATION;
+            break;
         case MKTAG('X', 'M', 'P', ' '):
             AV_WL32(chunk_str, chunk_type);
             av_log(avctx, AV_LOG_WARNING, "skipping unsupported chunk: %s\n",
@@ -1523,22 +1596,538 @@  exif_end:
         }
     }
 
-    if (!*got_frame) {
-        av_log(avctx, AV_LOG_ERROR, "image data not found\n");
-        return AVERROR_INVALIDDATA;
+    return size;
+}
+
+int init_canvas_frame(WebPContext *s, int format, int key_frame)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int height;
+    int ret;
+
+    // canvas is needed only for animation
+    if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION))
+        return 0;
+
+    // avoid init for non-key frames whose format and size did not change
+    if (!key_frame &&
+        canvas->data[0] &&
+        canvas->format == format &&
+        canvas->width  == s->canvas_width &&
+        canvas->height == s->canvas_height)
+        return 0;
+
+    // canvas changes within IPPP sequences will loose thread sync
+    // because of the ThreadFrame reallocation and will wait forever
+    // so if frame-threading is used, forbid canvas changes and unlock
+    // previous frames
+    if (!key_frame && canvas->data[0]) {
+        if (s->avctx->thread_count > 1) {
+            av_log(s->avctx, AV_LOG_WARNING, "Canvas change detected. The output will be damaged. Use -threads 1 to try decoding with best effort.\n");
+            // unlock previous frames that have sent an _await() call
+            ff_thread_report_progress(&s->canvas_frame, INT_MAX, 0);
+            return AVERROR_PATCHWELCOME;
+        } else {
+            // warn for damaged frames
+            av_log(s->avctx, AV_LOG_WARNING, "Canvas change detected. The output will be damaged.\n");
+        }
+    }
+
+    s->avctx->pix_fmt = format;
+    canvas->format    = format;
+    canvas->width     = s->canvas_width;
+    canvas->height    = s->canvas_height;
+
+    // VP8 decoder changed the width and height in AVCodecContext.
+    // Change it back to the canvas size.
+    ret = ff_set_dimensions(s->avctx, s->canvas_width, s->canvas_height);
+    if (ret < 0)
+        return ret;
+
+    ff_thread_release_ext_buffer(s->avctx, &s->canvas_frame);
+    ret = ff_thread_get_ext_buffer(s->avctx, &s->canvas_frame, AV_GET_BUFFER_FLAG_REF);
+    if (ret < 0)
+        return ret;
+
+    if (canvas->format == AV_PIX_FMT_ARGB) {
+        height = canvas->height;
+        memset(canvas->data[0], 0, height * canvas->linesize[0]);
+    } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+        const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(canvas->format);
+        for (int comp = 0; comp < desc->nb_components; comp++) {
+            int plane = desc->comp[comp].plane;
+
+            if (comp == 1 || comp == 2)
+                height = AV_CEIL_RSHIFT(canvas->height, desc->log2_chroma_h);
+            else
+                height = FFALIGN(canvas->height, 1 << desc->log2_chroma_h);
+
+            memset(canvas->data[plane], s->transparent_yuva[plane],
+                   height * canvas->linesize[plane]);
+        }
+    }
+
+    return 0;
+}
+
+/*
+ * Blend src1 (foreground) and src2 (background) into dest, in ARGB format.
+ * width, height are the dimensions of src1
+ * pos_x, pos_y is the position in src2 and in dest
+ */
+static void blend_alpha_argb(uint8_t *dest_data[4], int dest_linesize[4],
+                             const uint8_t *src1_data[4], int src1_linesize[4],
+                             const uint8_t *src2_data[4], int src2_linesize[4],
+                             int src2_step[4],
+                             int width, int height, int pos_x, int pos_y)
+{
+    for (int y = 0; y < height; y++) {
+        const uint8_t *src1 = src1_data[0] + y * src1_linesize[0];
+        const uint8_t *src2 = src2_data[0] + (y + pos_y) * src2_linesize[0] + pos_x * src2_step[0];
+        uint8_t       *dest = dest_data[0] + (y + pos_y) * dest_linesize[0] + pos_x * sizeof(uint32_t);
+        for (int x = 0; x < width; x++) {
+            int src1_alpha = src1[0];
+            int src2_alpha = src2[0];
+
+            if (src1_alpha == 255) {
+                memcpy(dest, src1, sizeof(uint32_t));
+            } else if (src1_alpha + src2_alpha == 0) {
+                memset(dest, 0, sizeof(uint32_t));
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+
+                dest[0] = blend_alpha;
+                dest[1] = ROUNDED_DIV(src1[1] * src1_alpha + src2[1] * tmp_alpha, blend_alpha);
+                dest[2] = ROUNDED_DIV(src1[2] * src1_alpha + src2[2] * tmp_alpha, blend_alpha);
+                dest[3] = ROUNDED_DIV(src1[3] * src1_alpha + src2[3] * tmp_alpha, blend_alpha);
+            }
+            src1 += sizeof(uint32_t);
+            src2 += src2_step[0];
+            dest += sizeof(uint32_t);
+        }
+    }
+}
+
+/*
+ * Blend src1 (foreground) and src2 (background) into dest, in YUVA format.
+ * width, height are the dimensions of src1
+ * pos_x, pos_y is the position in src2 and in dest
+ */
+static void blend_alpha_yuva(WebPContext *s,
+                             uint8_t *dest_data[4], int dest_linesize[4],
+                             const uint8_t *src1_data[4], int src1_linesize[4],
+                             int src1_format,
+                             const uint8_t *src2_data[4], int src2_linesize[4],
+                             int src2_step[4],
+                             int width, int height, int pos_x, int pos_y)
+{
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(src1_format);
+
+    int plane_y = desc->comp[0].plane;
+    int plane_u = desc->comp[1].plane;
+    int plane_v = desc->comp[2].plane;
+    int plane_a = desc->comp[3].plane;
+
+    // blend U & V planes first, because the later step may modify alpha plane
+    int w  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+    int h  = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+    int px = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+    int py = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+    int tile_w = 1 << desc->log2_chroma_w;
+    int tile_h = 1 << desc->log2_chroma_h;
+
+    for (int y = 0; y < h; y++) {
+        const uint8_t *src1_u = src1_data[plane_u] + y * src1_linesize[plane_u];
+        const uint8_t *src1_v = src1_data[plane_v] + y * src1_linesize[plane_v];
+        const uint8_t *src2_u = src2_data[plane_u] + (y + py) * src2_linesize[plane_u] + px * src2_step[plane_u];
+        const uint8_t *src2_v = src2_data[plane_v] + (y + py) * src2_linesize[plane_v] + px * src2_step[plane_v];
+        uint8_t       *dest_u = dest_data[plane_u] + (y + py) * dest_linesize[plane_u] + px;
+        uint8_t       *dest_v = dest_data[plane_v] + (y + py) * dest_linesize[plane_v] + px;
+        for (int x = 0; x < w; x++) {
+            // calculate the average alpha of the tile
+            int src1_alpha = 0;
+            int src2_alpha = 0;
+            for (int yy = 0; yy < tile_h; yy++) {
+                for (int xx = 0; xx < tile_w; xx++) {
+                    src1_alpha += src1_data[plane_a][(y * tile_h + yy) * src1_linesize[plane_a] +
+                                                     (x * tile_w + xx)];
+                    src2_alpha += src2_data[plane_a][((y + py) * tile_h + yy) * src2_linesize[plane_a] +
+                                                     ((x + px) * tile_w + xx) * src2_step[plane_a]];
+                }
+            }
+            src1_alpha = AV_CEIL_RSHIFT(src1_alpha, desc->log2_chroma_w + desc->log2_chroma_h);
+            src2_alpha = AV_CEIL_RSHIFT(src2_alpha, desc->log2_chroma_w + desc->log2_chroma_h);
+
+            if (src1_alpha == 255) {
+                *dest_u = *src1_u;
+                *dest_v = *src1_v;
+            } else if (src1_alpha + src2_alpha == 0) {
+                *dest_u = s->transparent_yuva[plane_u];
+                *dest_v = s->transparent_yuva[plane_v];
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+                *dest_u = ROUNDED_DIV(*src1_u * src1_alpha + *src2_u * tmp_alpha, blend_alpha);
+                *dest_v = ROUNDED_DIV(*src1_v * src1_alpha + *src2_v * tmp_alpha, blend_alpha);
+            }
+            src1_u++;
+            src1_v++;
+            src2_u += src2_step[plane_u];
+            src2_v += src2_step[plane_v];
+            dest_u++;
+            dest_v++;
+        }
+    }
+
+    // blend Y & A planes
+    for (int y = 0; y < height; y++) {
+        const uint8_t *src1_y = src1_data[plane_y] + y * src1_linesize[plane_y];
+        const uint8_t *src1_a = src1_data[plane_a] + y * src1_linesize[plane_a];
+        const uint8_t *src2_y = src2_data[plane_y] + (y + pos_y) * src2_linesize[plane_y] + pos_x * src2_step[plane_y];
+        const uint8_t *src2_a = src2_data[plane_a] + (y + pos_y) * src2_linesize[plane_a] + pos_x * src2_step[plane_a];
+        uint8_t       *dest_y = dest_data[plane_y] + (y + pos_y) * dest_linesize[plane_y] + pos_x;
+        uint8_t       *dest_a = dest_data[plane_a] + (y + pos_y) * dest_linesize[plane_a] + pos_x;
+        for (int x = 0; x < width; x++) {
+            int src1_alpha = *src1_a;
+            int src2_alpha = *src2_a;
+
+            if (src1_alpha == 255) {
+                *dest_y = *src1_y;
+                *dest_a = 255;
+            } else if (src1_alpha + src2_alpha == 0) {
+                *dest_y = s->transparent_yuva[plane_y];
+                *dest_a = 0;
+            } else {
+                int tmp_alpha = src2_alpha - ROUNDED_DIV(src1_alpha * src2_alpha, 255);
+                int blend_alpha = src1_alpha + tmp_alpha;
+                *dest_y = ROUNDED_DIV(*src1_y * src1_alpha + *src2_y * tmp_alpha, blend_alpha);
+                *dest_a = blend_alpha;
+            }
+            src1_y++;
+            src1_a++;
+            src2_y += src2_step[plane_y];
+            src2_a += src2_step[plane_a];
+            dest_y++;
+            dest_a++;
+        }
+    }
+}
+
+static int blend_frame_into_canvas(WebPContext *s)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    AVFrame *frame  = s->frame;
+    int width, height;
+    int pos_x, pos_y;
+
+    if ((s->anmf_flags & ANMF_BLENDING_METHOD) == ANMF_BLENDING_METHOD_OVERWRITE
+        || frame->format == AV_PIX_FMT_YUV420P) {
+        // do not blend, overwrite
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            width  = s->width;
+            height = s->height;
+            pos_x  = s->pos_x;
+            pos_y  = s->pos_y;
+
+            for (int y = 0; y < height; y++) {
+                const uint32_t *src = (uint32_t *) (frame->data[0] + y * frame->linesize[0]);
+                uint32_t *dst = (uint32_t *) (canvas->data[0] + (y + pos_y) * canvas->linesize[0]) + pos_x;
+                memcpy(dst, src, width * sizeof(uint32_t));
+            }
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(frame->format);
+            int plane;
+
+            for (int comp = 0; comp < desc->nb_components; comp++) {
+                plane  = desc->comp[comp].plane;
+                width  = s->width;
+                height = s->height;
+                pos_x  = s->pos_x;
+                pos_y  = s->pos_y;
+                if (comp == 1 || comp == 2) {
+                    width  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+                    height = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+                    pos_x  = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+                    pos_y  = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+                }
+
+                for (int y = 0; y < height; y++) {
+                    const uint8_t *src = frame->data[plane] + y * frame->linesize[plane];
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memcpy(dst, src, width);
+                }
+            }
+
+            if (desc->nb_components < 4) {
+                // frame does not have alpha, set alpha to 255
+                desc = av_pix_fmt_desc_get(canvas->format);
+                plane  = desc->comp[3].plane;
+                width  = s->width;
+                height = s->height;
+                pos_x  = s->pos_x;
+                pos_y  = s->pos_y;
+
+                for (int y = 0; y < height; y++) {
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memset(dst, 255, width);
+                }
+            }
+        }
+    } else {
+        // alpha blending
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            int src2_step[4] = { sizeof(uint32_t) };
+            blend_alpha_argb(canvas->data, canvas->linesize,
+                             (const uint8_t **) frame->data, frame->linesize,
+                             (const uint8_t **) canvas->data, canvas->linesize,
+                             src2_step, s->width, s->height, s->pos_x, s->pos_y);
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            int src2_step[4] = { 1, 1, 1, 1 };
+            blend_alpha_yuva(s, canvas->data, canvas->linesize,
+                             (const uint8_t **) frame->data, frame->linesize,
+                             frame->format,
+                             (const uint8_t **) canvas->data, canvas->linesize,
+                             src2_step, s->width, s->height, s->pos_x, s->pos_y);
+        }
+    }
+
+    return 0;
+}
+
+static int copy_canvas_to_frame(WebPContext *s, AVFrame *frame, int key_frame)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int ret;
+
+    frame->format = canvas->format;
+    frame->width  = canvas->width;
+    frame->height = canvas->height;
+
+    ret = av_frame_get_buffer(frame, 0);
+    if (ret < 0)
+        return ret;
+
+    ret = av_frame_copy_props(frame, canvas);
+    if (ret < 0)
+        return ret;
+
+    ret = ff_attach_decode_data(frame);
+    if (ret < 0)
+        return ret;
+
+    // blend the canvas with the background color into the output frame
+    if (canvas->format == AV_PIX_FMT_ARGB) {
+        int src2_step[4] = { 0 };
+        const uint8_t *src2_data[4] = { &s->background_argb[0] };
+        blend_alpha_argb(frame->data, frame->linesize,
+                         (const uint8_t **) canvas->data, canvas->linesize,
+                         (const uint8_t **) src2_data, src2_step, src2_step,
+                         canvas->width, canvas->height, 0, 0);
+    } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+        int src2_step[4] = { 0, 0, 0, 0 };
+        blend_alpha_yuva(s, frame->data, frame->linesize,
+                         (const uint8_t **) canvas->data, canvas->linesize,
+                         canvas->format,
+                         s->background_data, src2_step, src2_step,
+                         canvas->width, canvas->height, 0, 0);
+    }
+
+    if (key_frame) {
+        frame->pict_type = AV_PICTURE_TYPE_I;
+    } else {
+        frame->pict_type = AV_PICTURE_TYPE_P;
+    }
+
+    return 0;
+}
+
+static int dispose_prev_frame_in_canvas(WebPContext *s)
+{
+    AVFrame *canvas = s->canvas_frame.f;
+    int width, height;
+    int pos_x, pos_y;
+
+    if ((s->prev_anmf_flags & ANMF_DISPOSAL_METHOD) == ANMF_DISPOSAL_METHOD_BACKGROUND) {
+        // dispose to background
+
+        if (canvas->format == AV_PIX_FMT_ARGB) {
+            width  = s->prev_width;
+            height = s->prev_height;
+            pos_x  = s->prev_pos_x;
+            pos_y  = s->prev_pos_y;
+
+            for (int y = 0; y < height; y++) {
+                uint32_t *dst = (uint32_t *) (canvas->data[0] + (y + pos_y) * canvas->linesize[0]) + pos_x;
+                memset(dst, 0, width * sizeof(uint32_t));
+            }
+        } else /* if (canvas->format == AV_PIX_FMT_YUVA420P) */ {
+            const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(canvas->format);
+            int plane;
+
+            for (int comp = 0; comp < desc->nb_components; comp++) {
+                plane  = desc->comp[comp].plane;
+                width  = s->prev_width;
+                height = s->prev_height;
+                pos_x  = s->prev_pos_x;
+                pos_y  = s->prev_pos_y;
+                if (comp == 1 || comp == 2) {
+                    width  = AV_CEIL_RSHIFT(width,  desc->log2_chroma_w);
+                    height = AV_CEIL_RSHIFT(height, desc->log2_chroma_h);
+                    pos_x  = AV_CEIL_RSHIFT(pos_x,  desc->log2_chroma_w);
+                    pos_y  = AV_CEIL_RSHIFT(pos_y,  desc->log2_chroma_h);
+                }
+
+                for (int y = 0; y < height; y++) {
+                    uint8_t *dst = canvas->data[plane] + (y + pos_y) * canvas->linesize[plane] + pos_x;
+                    memset(dst, s->transparent_yuva[plane], width);
+                }
+            }
+        }
+    }
+
+    return 0;
+}
+
+static int webp_decode_frame(AVCodecContext *avctx, AVFrame *p,
+                             int *got_frame, AVPacket *avpkt)
+{
+    WebPContext *s = avctx->priv_data;
+    AVFrame *canvas = s->canvas_frame.f;
+    int ret;
+    int key_frame = avpkt->flags & AV_PKT_FLAG_KEY;
+
+    for (int i = 0; i < avpkt->side_data_elems; ++i) {
+        if (avpkt->side_data[i].type == AV_PKT_DATA_NEW_EXTRADATA) {
+            ret = webp_decode_frame_common(avctx, avpkt->side_data[i].data,
+                                           avpkt->side_data[i].size,
+                                           got_frame, key_frame);
+            if (ret < 0)
+                goto end;
+        }
+    }
+
+    *got_frame   = 0;
+
+    if (key_frame) {
+        // The canvas is passed from one thread to another in a sequence
+        // starting with a key frame followed by non-key frames.
+        // The key frame reports progress 1,
+        // the N-th non-key frame awaits progress N = s->await_progress
+        // and reports progress N + 1.
+        s->await_progress = 0;
+    }
+
+    // reset the frame params
+    s->anmf_flags = 0;
+    s->width      = 0;
+    s->height     = 0;
+    s->pos_x      = 0;
+    s->pos_y      = 0;
+    s->has_alpha  = 0;
+
+    ret = webp_decode_frame_common(avctx, avpkt->data, avpkt->size, got_frame, key_frame);
+    if (ret < 0)
+        goto end;
+
+    if (s->vp8x_flags & VP8X_FLAG_ANIMATION) {
+            // no animation, output the decoded frame
+        // VP8 decoder might have changed the width and height of the frame
+        AVFrame *frame  = s->frame;
+        ret = av_frame_copy_props(canvas, frame);
+        if (ret < 0)
+            return ret;
+
+        ret = ff_set_dimensions(s->avctx, canvas->width, canvas->height);
+        if (ret < 0)
+            return ret;
+
+        s->avctx->pix_fmt = canvas->format;
+    }
+
+    ff_thread_finish_setup(s->avctx);
+
+    if (*got_frame) {
+        if (!(s->vp8x_flags & VP8X_FLAG_ANIMATION)) {
+            // no animation, output the decoded frame
+            av_frame_move_ref(p, s->frame);
+            ret = ff_attach_decode_data(p);
+            if (ret < 0)
+                return ret;
+        } else {
+            if (!key_frame) {
+                ff_thread_await_progress(&s->canvas_frame, s->await_progress, 0);
+
+                ret = dispose_prev_frame_in_canvas(s);
+                if (ret < 0)
+                    goto end;
+            }
+
+            ret = blend_frame_into_canvas(s);
+            if (ret < 0)
+                goto end;
+
+            ret = copy_canvas_to_frame(s, p, key_frame);
+            if (ret < 0)
+                goto end;
+
+            ff_thread_report_progress(&s->canvas_frame, s->await_progress + 1, 0);
+        }
+
+        p->pts = avpkt->pts;
     }
 
-    return avpkt->size;
+    ret = avpkt->size;
+
+end:
+    av_frame_unref(s->frame);
+    return ret;
 }
 
 static av_cold int webp_decode_init(AVCodecContext *avctx)
 {
     WebPContext *s = avctx->priv_data;
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(AV_PIX_FMT_YUVA420P);
+    int ret;
+    const AVCodec *codec;
 
+    s->avctx = avctx;
     s->pkt = av_packet_alloc();
-    if (!s->pkt)
+    s->canvas_frame.f = av_frame_alloc();
+    s->frame = av_frame_alloc();
+    if (!s->pkt || !s->canvas_frame.f || !s->frame) {
+        av_packet_free(&s->pkt);
+        av_frame_free(&s->canvas_frame.f);
+        av_frame_free(&s->frame);
         return AVERROR(ENOMEM);
+    }
 
+    // prepare data pointers for YUVA background
+    for (int i = 0; i < 4; i++)
+        s->background_data[i] = &s->background_yuva[i];
+
+    // convert transparent black from RGBA to YUVA
+    s->transparent_yuva[desc->comp[0].plane] = RGB_TO_Y_CCIR(0, 0, 0);
+    s->transparent_yuva[desc->comp[1].plane] = RGB_TO_U_CCIR(0, 0, 0, 0);
+    s->transparent_yuva[desc->comp[2].plane] = RGB_TO_V_CCIR(0, 0, 0, 0);
+    s->transparent_yuva[desc->comp[3].plane] = 0;
+
+    /* Prepare everything needed for VP8 decoding */
+    codec = avcodec_find_decoder(AV_CODEC_ID_VP8);
+    if (!codec)
+        return AVERROR_BUG;
+    s->avctx_vp8 = avcodec_alloc_context3(codec);
+    if (!s->avctx_vp8)
+        return AVERROR(ENOMEM);
+    s->avctx_vp8->flags = avctx->flags;
+    s->avctx_vp8->flags2 = avctx->flags2;
+    s->avctx_vp8->pix_fmt = avctx->pix_fmt;
+    ret = avcodec_open2(s->avctx_vp8, codec, NULL);
+    if (ret < 0) {
+        return ret;
+    }
     return 0;
 }
 
@@ -1547,12 +2136,52 @@  static av_cold int webp_decode_close(AVCodecContext *avctx)
     WebPContext *s = avctx->priv_data;
 
     av_packet_free(&s->pkt);
+    ff_thread_release_ext_buffer(&s->canvas_frame);
+    av_frame_free(&s->canvas_frame.f);
+    av_frame_free(&s->frame);
+    avcodec_free_context(&s->avctx_vp8);
+
+    return 0;
+}
+
+static void webp_decode_flush(AVCodecContext *avctx)
+{
+    WebPContext *s = avctx->priv_data;
+
+    ff_thread_release_ext_buffer(&s->canvas_frame);
+}
+
+#if HAVE_THREADS
+static int webp_update_thread_context(AVCodecContext *dst, const AVCodecContext *src)
+{
+    WebPContext *wsrc = src->priv_data;
+    WebPContext *wdst = dst->priv_data;
+    int ret;
+
+    if (dst == src)
+        return 0;
+
+    ff_thread_release_ext_buffer(&wdst->canvas_frame);
+    if (wsrc->canvas_frame.f->data[0] &&
+        (ret = ff_thread_ref_frame(&wdst->canvas_frame, &wsrc->canvas_frame)) < 0)
+        return ret;
+
+    wdst->vp8x_flags      = wsrc->vp8x_flags;
+    wdst->canvas_width    = wsrc->canvas_width;
+    wdst->canvas_height   = wsrc->canvas_height;
+    wdst->prev_anmf_flags = wsrc->anmf_flags;
+    wdst->prev_width      = wsrc->width;
+    wdst->prev_height     = wsrc->height;
+    wdst->prev_pos_x      = wsrc->pos_x;
+    wdst->prev_pos_y      = wsrc->pos_y;
+    wdst->await_progress  = wsrc->await_progress + 1;
 
-    if (s->initialized)
-        return ff_vp8_decode_free(avctx);
+    memcpy(wdst->background_argb,  wsrc->background_argb,  sizeof(wsrc->background_argb));
+    memcpy(wdst->background_yuva,  wsrc->background_yuva,  sizeof(wsrc->background_yuva));
 
     return 0;
 }
+#endif
 
 const FFCodec ff_webp_decoder = {
     .p.name         = "webp",
@@ -1560,9 +2189,11 @@  const FFCodec ff_webp_decoder = {
     .p.type         = AVMEDIA_TYPE_VIDEO,
     .p.id           = AV_CODEC_ID_WEBP,
     .priv_data_size = sizeof(WebPContext),
+    UPDATE_THREAD_CONTEXT(webp_update_thread_context),
     .init           = webp_decode_init,
     FF_CODEC_DECODE_CB(webp_decode_frame),
     .close          = webp_decode_close,
+    .flush          = webp_decode_flush,
     .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_FRAME_THREADS,
-    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES,
+    .caps_internal  = FF_CODEC_CAP_ICC_PROFILES | FF_CODEC_CAP_ALLOCATE_PROGRESS,
 };