Skip to content

Use ass_set_storage_size to compensate for target SAR #19

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
JustinTArthur wants to merge 2 commits into
base: master
Choose a base branch
from
Open

Use ass_set_storage_size to compensate for target SAR #19

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f3c0d38
Use ass_set_storage_size to compensate for target SAR.
JustinTArthur Aug 18, 2026
bccff88
Only adapt to SAR for tracks this plugin lays out
JustinTArthur Aug 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,6 +60,20 @@ Subtext is a subtitle renderer that uses libass and ffmpeg.
worth doing when blend=False and producing the frames of *clip*
is expensive.

Only subtitles this plugin lays out are followed that way:
the text given to Subtitle, and formats converted to ASS such
as SRT and WebVTT. An ASS file is left exactly as its author
wrote it, the way other renderers treat it, so its subtitles
are squeezed along with the picture. Setting *sar* asks for
the file to be adapted anyway.

Any compensation only rescales subtitles horizontally, keeping the
relationship between font size and height constant.

Subtitles carrying ``LayoutResX`` and ``LayoutResY`` headers
already express the shape they're laid out for and are rendered to it.
In this case, neither frame properties nor this parameter are used.

style
Custom ASS style for subtitle formats other than ASS. If empty
(the default), libavcodec's default style is used. This
Expand Down
84 changes: 71 additions & 13 deletions src/text.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@
*/

#include <cinttypes>
#include <cmath>
#include <cstdarg>
#include <cstdio>
#include <cstring>
Expand Down Expand Up @@ -56,6 +57,7 @@ struct AssData {
VSFrame *lastframe = nullptr;

bool adapt_sar = false;
bool own_track = false;

AssLibraryPtr ass_library;
AssRendererPtr ass_renderer;
Expand Down Expand Up @@ -151,9 +153,9 @@ static void assRender(VSFrame *dst, VSFrame *alpha, const VSAPI *vsapi,
}
}

/* Reads the pixel aspect ratio out of the frame properties of the clip.
/* Reads the sample aspect ratio out of the frame properties of the clip.
Square pixels unless the frame says otherwise. */
static double frameToPixelAspect(const VSFrame *src, const VSAPI *vsapi)
static double frameToSampleAspect(const VSFrame *src, const VSAPI *vsapi)
{
const VSMap *props = vsapi->getFramePropertiesRO(src);
int num_err, den_err;
Expand All @@ -167,6 +169,49 @@ static double frameToPixelAspect(const VSFrame *src, const VSAPI *vsapi)
return (double)sar_num / (double)sar_den;
}

/* Tells libass what the subtitles are laid out for in square pixel
resolution: the width of the clip scaled by the SAR at the height the clip
already has.

Only width moves because libass divides width by display scale for blur. */
static void setLayoutSize(AssData *d, double sar)
{
int layout_x = (int)std::lround(d->vi[0].width * sar);

if(layout_x <= 0)
layout_x = d->vi[0].width;

ass_set_storage_size(d->ass_renderer.get(), layout_x, d->vi[0].height);

/* PlayRes is the coordinate space the subtitles are written in, and libass
measures drawings, letter spacing and borders against it, so it has to
describe the same display shape the layout resolution does or only the
glyphs come out compensated. The vertical half is left alone because
font sizes are relative to it; only PlayResX moves. Tracks read from a
subtitle file say what they were written for and are never touched. */
if(d->own_track && d->ass && d->ass->PlayResY > 0) {
int play_x = (int)std::lround((double)d->ass->PlayResY *
layout_x / d->vi[0].height);

if(play_x > 0 && play_x != d->ass->PlayResX) {
/* Style margins are measured in the same units, so they have to
follow PlayResX to keep covering the same part of the frame. */
if(d->ass->PlayResX > 0) {
double fix = (double)play_x / d->ass->PlayResX;

for(int i = 0; i < d->ass->n_styles; i++) {
d->ass->styles[i].MarginL =
(int)std::lround(d->ass->styles[i].MarginL * fix);
d->ass->styles[i].MarginR =
(int)std::lround(d->ass->styles[i].MarginR * fix);
}
}

d->ass->PlayResX = play_x;
}
}
}

static const VSFrame *VS_CC assGetFrame(int n, int activationReason,
void *instanceData, [[maybe_unused]] void **frameData,
VSFrameContext *frameCtx, VSCore *core,
Expand Down Expand Up @@ -194,14 +239,14 @@ static const VSFrame *VS_CC assGetFrame(int n, int activationReason,

if(d->adapt_sar) {
const VSFrame *src = vsapi->getFrameFilter(n, d->node, frameCtx);
double par = frameToPixelAspect(src, vsapi);
double sar = frameToSampleAspect(src, vsapi);
vsapi->freeFrame(src);

/* Setting the same value again costs nothing, libass only
reconfigures when the aspect ratio actually changes. A change
empties its caches, which makes the rendered images differ, so
the change detection below picks it up. */
ass_set_pixel_aspect(d->ass_renderer.get(), par);
/* Setting the same value again costs nothing, the layout only
changes when the aspect ratio does. A change empties the libass
caches, which makes the rendered images differ, so the change
detection below picks it up. */
setLayoutSize(d, sar);
}

ts = (int64_t)n * 1000 * d->vi[0].fpsDen / d->vi[0].fpsNum;
Expand Down Expand Up @@ -412,17 +457,13 @@ static void VS_CC assRenderCreate(const VSMap *in, VSMap *out, void *userData,

ass_set_font_scale(d.ass_renderer.get(), scale);
ass_set_frame_size(d.ass_renderer.get(), d.vi[0].width, d.vi[0].height);
ass_set_storage_size(d.ass_renderer.get(), d.vi[0].width, d.vi[0].height);
ass_set_margins(d.ass_renderer.get(),
margins[0], margins[1], margins[2], margins[3]);
ass_set_use_margins(d.ass_renderer.get(), 0);

if(linespacing)
ass_set_line_spacing(d.ass_renderer.get(), linespacing);

if(sar)
ass_set_pixel_aspect(d.ass_renderer.get(), sar);

if(fontdir)
ass_set_fonts_dir(d.ass_library.get(), fontdir);

Expand All @@ -446,6 +487,7 @@ static void VS_CC assRenderCreate(const VSMap *in, VSMap *out, void *userData,
"ScriptType: v4.00+\n"
"PlayResX: " + std::to_string(d.vi[0].width) + "\n"
"PlayResY: " + std::to_string(d.vi[0].height) + "\n"
"ScaledBorderAndShadow: yes\n"
"[V4+ Styles]\n"
"Format: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding\n"
"Style: Default," + *style + "\n"
Expand All @@ -461,15 +503,21 @@ static void VS_CC assRenderCreate(const VSMap *in, VSMap *out, void *userData,
}

ass_process_data(d.ass.get(), script.data(), (int)script.size());

d.own_track = true;
} else {
std::string conversion_error;
std::optional<std::string> contents = convertToUtf8(*file, charset, conversion_error);

if (contents) {
d.ass.reset(ass_read_memory(d.ass_library.get(), contents->data(), contents->size(), NULL));

if (!d.ass)
if (!d.ass) {
d.ass.reset(convertToASS(*file, *contents, d.ass_library.get(), style ? style->c_str() : nullptr, charset, conversion_error));

/* libavcodec's conversion, not a script anybody authored. */
d.own_track = true;
}
}

if (!d.ass) {
Expand All @@ -478,6 +526,16 @@ static void VS_CC assRenderCreate(const VSMap *in, VSMap *out, void *userData,
}
}

/* Subtitles read from a file were written for the video the way it is
stored, which is how every other renderer treats them, so the aspect of
the clip is only followed for the tracks made here. Passing sar asks for
it anyway. A track carrying LayoutRes headers says what it's laid out
for and wins over both, no rendered frame needed. */
if(!d.own_track || (d.ass->LayoutResX > 0 && d.ass->LayoutResY > 0))
d.adapt_sar = false;

setLayoutSize(&d, sar ? sar : 1);

VSFrame *frame = vsapi->newVideoFrame(&d.vi[0].format,
d.vi[0].width,
d.vi[0].height,
Expand Down