优化播放

internal/service/service_builder.go

@@ -124,6 +124,7 @@ func (b *serviceContainerBuilder) initContentServices() {
 	// STRM 直连失败后的 HLS 转码：把 .strm 解析成 ffmpeg 可读取的本地路径或 HTTP 直链。
 	b.c.Transcoder.SetStrmPlayTargetResolver(b.c.Strm.ResolvePlayTarget)
 	b.c.Transcoder.SetProbe(b.c.FFprobe)
+	b.c.Subtitle.SetStrmPlayTargetResolver(b.c.Strm.ResolvePlayTarget)
 	// 弹幕识别需要把远程 Emby 条目解析为 Media 元数据及可拉取前 16MB 的直链 URL。
 	if b.c.EmbyRemote != nil {
 		b.c.Danmaku.SetRemoteMediaResolver(func(ctx context.Context, encodedID string) (*model.Media, string, error) {

internal/service/stream_hls.go

@@ -25,7 +25,8 @@ func (s *StreamService) ServeHLSPlaylist(w http.ResponseWriter, r *http.Request,
 	}
 	startSec := parseHLSStartSec(r)
 	seekGen := parseHLSSeekGen(r)
-	if _, err := s.transcoder.EnsureJobFrom(r.Context(), mediaID, startSec, seekGen); err != nil {
+	subtitleStream := parseHLSSubtitleStream(r)
+	if _, err := s.transcoder.EnsureJobFromSubtitle(r.Context(), mediaID, startSec, seekGen, subtitleStream); err != nil {
 		return err
 	}
 	s.transcoder.TouchJob(mediaID)
@@ -60,6 +61,21 @@ func (s *StreamService) ServeHLSPlaylist(w http.ResponseWriter, r *http.Request,
 	return nil
 }
 
+func parseHLSSubtitleStream(r *http.Request) int {
+	if r == nil {
+		return -1
+	}
+	raw := strings.TrimSpace(r.URL.Query().Get("subtitle"))
+	if raw == "" {
+		return -1
+	}
+	v, err := strconv.Atoi(raw)
+	if err != nil || v < 0 {
+		return -1
+	}
+	return v
+}
+
 func parseHLSStartSec(r *http.Request) float64 {
 	if r == nil {
 		return 0

internal/service/subtitle.go

@@ -16,12 +16,15 @@ package service
 
 import (
 	"context"
+	"encoding/json"
 	"errors"
 	"fmt"
 	"io"
 	"os"
+	"os/exec"
 	"path/filepath"
 	"regexp"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -29,14 +32,16 @@ import (
 	"go.uber.org/zap"
 
 	"github.com/truewhile/MeBox/internal/config"
+	"github.com/truewhile/MeBox/internal/model"
 	"github.com/truewhile/MeBox/internal/repository"
 )
 
 // SubtitleService is the discovery + conversion entry point.
 type SubtitleService struct {
-	log  *zap.Logger
-	repo *repository.Container
-	cfg  *config.Config
+	log         *zap.Logger
+	repo        *repository.Container
+	cfg         *config.Config
+	strmResolve func(ctx context.Context, raw string) (*StrmPlayResult, error)
 
 	// 目录发现是 Emby 条目列表的热路径（每个媒体源一次 DB 查询 + 最多 5 次
 	// os.ReadDir），而字幕文件极少变化：按 media_id 做短 TTL 缓存。
@@ -61,11 +66,14 @@ func NewSubtitleService(cfg *config.Config, log *zap.Logger, repo *repository.Co
 
 // SubtitleTrack describes one external subtitle file.
 type SubtitleTrack struct {
-	Lang  string `json:"lang"`
-	Label string `json:"label"`
-	Path  string `json:"path"`
-	URL   string `json:"url"`
-	Codec string `json:"codec"`
+	Lang        string `json:"lang"`
+	Label       string `json:"label"`
+	Path        string `json:"path"`
+	URL         string `json:"url"`
+	Codec       string `json:"codec"`
+	Source      string `json:"source"`
+	Delivery    string `json:"delivery"`
+	StreamIndex int    `json:"stream_index,omitempty"`
 }
 
 // extToCodec maps the file extension to the inner codec name.
@@ -86,7 +94,17 @@ func (s *SubtitleService) Discover(ctx context.Context, mediaID string) ([]Subti
 // DiscoverExternalOnly 只返回媒体旁边的外挂字幕文件，不含容器内嵌字幕轨。
 // Emby 字幕接口（/Videos/:id/Subtitles/...）用。
 func (s *SubtitleService) DiscoverExternalOnly(ctx context.Context, mediaID string) ([]SubtitleTrack, error) {
-	return s.discover(ctx, mediaID)
+	tracks, err := s.discover(ctx, mediaID)
+	if err != nil {
+		return nil, err
+	}
+	out := make([]SubtitleTrack, 0, len(tracks))
+	for _, track := range tracks {
+		if track.Source != "embedded" {
+			out = append(out, track)
+		}
+	}
+	return out, nil
 }
 
 func (s *SubtitleService) discover(ctx context.Context, mediaID string) ([]SubtitleTrack, error) {
@@ -187,16 +205,153 @@ func (s *SubtitleService) discoverUncached(ctx context.Context, mediaID string)
 			}
 			lang := detectLang(fullName, matchedBase)
 			tracks = append(tracks, SubtitleTrack{
-				Lang:  lang,
-				Label: lang,
-				Path:  filepath.Join(c, e.Name()),
-				Codec: codec,
+				Lang:     lang,
+				Label:    lang,
+				Path:     filepath.Join(c, e.Name()),
+				Codec:    codec,
+				Source:   "external",
+				Delivery: "webvtt",
 			})
 		}
 	}
+	embedded, err := s.discoverEmbedded(ctx, m)
+	if err != nil {
+		if s.log != nil {
+			s.log.Debug("discover embedded subtitles failed", zap.String("media_id", mediaID), zap.Error(err))
+		}
+	} else {
+		tracks = append(tracks, embedded...)
+	}
 	return tracks, nil
 }
 
+type embeddedSubtitleProbe struct {
+	Streams []struct {
+		Index     int    `json:"index"`
+		CodecName string `json:"codec_name"`
+		Tags      struct {
+			Language string `json:"language"`
+			Title    string `json:"title"`
+		} `json:"tags"`
+		Disposition struct {
+			Default int `json:"default"`
+			Forced  int `json:"forced"`
+		} `json:"disposition"`
+	} `json:"streams"`
+}
+
+var imageSubtitleCodecs = map[string]bool{
+	"hdmv_pgs_subtitle": true,
+	"dvd_subtitle":      true,
+	"dvb_subtitle":      true,
+	"xsub":              true,
+}
+
+func (s *SubtitleService) discoverEmbedded(ctx context.Context, media *model.Media) ([]SubtitleTrack, error) {
+	if s == nil || s.cfg == nil {
+		return nil, errors.New("subtitle probe unavailable")
+	}
+	input, err := s.resolveInput(ctx, media)
+	if err != nil {
+		return nil, err
+	}
+	bin, err := resolveLocalExecutable(s.cfg.App.FFprobePath, "ffprobe")
+	if err != nil {
+		return nil, err
+	}
+	probeCtx, cancel := context.WithTimeout(ctx, 30*time.Second)
+	defer cancel()
+	args := []string{"-v", "error"}
+	if headers := ffmpegHeaderText(input.Headers); headers != "" {
+		args = append(args, "-headers", headers)
+	}
+	args = append(args,
+		"-select_streams", "s",
+		"-show_entries", "stream=index,codec_name:stream_tags=language,title:stream_disposition=default,forced",
+		"-of", "json", input.Source,
+	)
+	out, err := exec.CommandContext(probeCtx, bin, args...).Output() // #nosec G204 -- executable is resolved locally and arguments do not use a shell.
+	if err != nil {
+		return nil, err
+	}
+	var probe embeddedSubtitleProbe
+	if err := json.Unmarshal(out, &probe); err != nil {
+		return nil, err
+	}
+	return subtitleTracksFromProbe(probe), nil
+}
+
+func subtitleTracksFromProbe(probe embeddedSubtitleProbe) []SubtitleTrack {
+	tracks := make([]SubtitleTrack, 0, len(probe.Streams))
+	for _, stream := range probe.Streams {
+		codec := strings.ToLower(strings.TrimSpace(stream.CodecName))
+		lang := strings.ToLower(strings.TrimSpace(stream.Tags.Language))
+		if lang == "" {
+			lang = "und"
+		}
+		label := strings.TrimSpace(stream.Tags.Title)
+		if label == "" {
+			label = lang
+		}
+		if stream.Disposition.Forced != 0 {
+			label += "（强制）"
+		} else if stream.Disposition.Default != 0 {
+			label += "（默认）"
+		}
+		delivery := "webvtt"
+		if imageSubtitleCodecs[codec] {
+			delivery = "burn"
+		}
+		sourceLabel := "（内嵌）"
+		if delivery == "burn" {
+			sourceLabel = "（内嵌·图片）"
+		}
+		tracks = append(tracks, SubtitleTrack{
+			Lang:        lang,
+			Label:       label + sourceLabel,
+			Path:        "embedded:" + strconv.Itoa(stream.Index),
+			Codec:       codec,
+			Source:      "embedded",
+			Delivery:    delivery,
+			StreamIndex: stream.Index,
+		})
+	}
+	return tracks
+}
+
+func (s *SubtitleService) SetStrmPlayTargetResolver(resolve func(context.Context, string) (*StrmPlayResult, error)) {
+	if s != nil {
+		s.strmResolve = resolve
+	}
+}
+
+func (s *SubtitleService) resolveInput(ctx context.Context, media *model.Media) (transcodeInput, error) {
+	if media == nil {
+		return transcodeInput{}, ErrMediaNotFound
+	}
+	if !isStrmMediaRow(media) {
+		if _, err := os.Stat(media.Path); err != nil {
+			return transcodeInput{}, ErrMediaNotFound
+		}
+		return transcodeInput{Source: media.Path}, nil
+	}
+	raw := strings.TrimSpace(media.STRMURL)
+	if raw == "" && strings.HasSuffix(strings.ToLower(media.Path), ".strm") {
+		raw, _ = readLocalSTRMTarget(media.Path)
+	}
+	if s.strmResolve != nil {
+		resolved, err := s.strmResolve(ctx, raw)
+		if err != nil {
+			return transcodeInput{}, err
+		}
+		return transcodeInputFromPlayResult(resolved)
+	}
+	if isHTTPPlaybackTarget(raw) {
+		return transcodeInput{Source: raw}, nil
+	}
+	return transcodeInput{}, errors.New("subtitle source unavailable")
+}
+
 // langTag matches the .zh / .zh-cn / .chs language sub-extensions.
 var langTag = regexp.MustCompile(`(?i)\.([a-z]{2,3}(?:[-_][a-z]{2,4})?)$`)
 
@@ -220,6 +375,13 @@ func (s *SubtitleService) Serve(ctx context.Context, mediaID, sub string, w io.W
 	if err != nil || m == nil {
 		return errors.New("media not found")
 	}
+	if strings.HasPrefix(sub, "embedded:") {
+		index, err := strconv.Atoi(strings.TrimPrefix(sub, "embedded:"))
+		if err != nil || index < 0 {
+			return errors.New("invalid embedded subtitle")
+		}
+		return s.serveEmbedded(ctx, m, index, w)
+	}
 	abs, err := filepath.Abs(sub)
 	if err != nil {
 		return err
@@ -252,6 +414,28 @@ func (s *SubtitleService) Serve(ctx context.Context, mediaID, sub string, w io.W
 	return err
 }
 
+func (s *SubtitleService) serveEmbedded(ctx context.Context, media *model.Media, streamIndex int, w io.Writer) error {
+	input, err := s.resolveInput(ctx, media)
+	if err != nil {
+		return err
+	}
+	bin, err := resolveLocalExecutable(s.cfg.App.FFmpegPath, "ffmpeg")
+	if err != nil {
+		return err
+	}
+	args := []string{"-hide_banner", "-loglevel", "error"}
+	args = append(args, ffmpegHTTPInputArgs(input)...)
+	args = append(args, "-i", input.Source, "-map", "0:"+strconv.Itoa(streamIndex), "-f", "webvtt", "-")
+	cmd := exec.CommandContext(ctx, bin, args...) // #nosec G204 -- executable is resolved locally and arguments do not use a shell.
+	cmd.Stdout = w
+	var stderr strings.Builder
+	cmd.Stderr = &stderr
+	if err := cmd.Run(); err != nil {
+		return fmt.Errorf("extract embedded subtitle: %w: %s", err, strings.TrimSpace(stderr.String()))
+	}
+	return nil
+}
+
 // ServeRaw writes the subtitle file in its original format without any
 // WebVTT conversion. Emby/Jellyfin clients advertise the source codec (ASS,
 // subrip, etc.) in MediaStreams, then fetch the subtitle bytes via the

internal/service/subtitle_test.go

@@ -2,6 +2,7 @@ package service
 
 import (
 	"bytes"
+	"encoding/json"
 	"os"
 	"path/filepath"
 	"testing"
@@ -46,6 +47,27 @@ func TestSubtitleDiscoverNoTracksReturnsEmptySlice(t *testing.T) {
 	}
 }
 
+func TestEmbeddedSubtitleProbeClassifiesTextAndBitmapTracks(t *testing.T) {
+	var probe embeddedSubtitleProbe
+	raw := []byte(`{"streams":[
+		{"index":2,"codec_name":"ass","tags":{"language":"chi","title":"中文"},"disposition":{"default":1}},
+		{"index":4,"codec_name":"hdmv_pgs_subtitle","tags":{"language":"eng"},"disposition":{"forced":1}}
+	]}`)
+	if err := json.Unmarshal(raw, &probe); err != nil {
+		t.Fatal(err)
+	}
+	tracks := subtitleTracksFromProbe(probe)
+	if len(tracks) != 2 {
+		t.Fatalf("len(tracks) = %d, want 2", len(tracks))
+	}
+	if tracks[0].Delivery != "webvtt" || tracks[0].Path != "embedded:2" {
+		t.Fatalf("text track = %#v", tracks[0])
+	}
+	if tracks[1].Delivery != "burn" || tracks[1].StreamIndex != 4 {
+		t.Fatalf("bitmap track = %#v", tracks[1])
+	}
+}
+
 func TestNormaliseTimecode(t *testing.T) {
 	cases := map[string]string{
 		"0:00:01":       "00:00:01",

internal/service/transcoder.go

@@ -71,6 +71,9 @@ type hlsJob struct {
 	// seekGen is the client `_seek` token for this job. Newer gens win;
 	// older/missing gens must not cancel a mid-file restart.
 	seekGen int64
+	// subtitleStream is an absolute ffmpeg stream index to burn into the
+	// picture, or -1 when no bitmap subtitle is selected.
+	subtitleStream int
 	// done is closed when the ffmpeg goroutine fully exits (after process death).
 	done chan struct{}
 }
@@ -121,6 +124,12 @@ func (t *TranscoderService) EnsureJob(ctx context.Context, mediaID string) (stri
 // in-flight playlist refreshes from a destroyed player commonly arrive as
 // start=0 right after a scrub and would otherwise reset playback to the head.
 func (t *TranscoderService) EnsureJobFrom(ctx context.Context, mediaID string, startSec float64, seekGen int64) (string, error) {
+	return t.EnsureJobFromSubtitle(ctx, mediaID, startSec, seekGen, -1)
+}
+
+// EnsureJobFromSubtitle is EnsureJobFrom with an optional bitmap subtitle
+// stream. Selecting or closing a burned subtitle creates a new HLS generation.
+func (t *TranscoderService) EnsureJobFromSubtitle(ctx context.Context, mediaID string, startSec float64, seekGen int64, subtitleStream int) (string, error) {
 	if !t.cfg.Transcoder.Enabled {
 		return "", ErrTranscodeDisabled
 	}
@@ -141,12 +150,12 @@ func (t *TranscoderService) EnsureJobFrom(ctx context.Context, mediaID string, s
 
 	t.mu.Lock()
 	if existing, ok := t.jobs[mediaID]; ok {
-		if sameHLSStart(existing.startSec, startSec) {
+		if sameHLSConfiguration(existing, startSec, subtitleStream) {
 			t.touchJobLocked(mediaID)
 			t.mu.Unlock()
 			return t.PlaylistPath(mediaID), nil
 		}
-		if !shouldReplaceHLSJob(existing, startSec, seekGen) {
+		if !shouldReplaceHLSJobConfiguration(existing, startSec, seekGen, subtitleStream) {
 			t.touchJobLocked(mediaID)
 			t.mu.Unlock()
 			return t.PlaylistPath(mediaID), nil
@@ -163,6 +172,9 @@ func (t *TranscoderService) EnsureJobFrom(ctx context.Context, mediaID string, s
 		return "", err
 	}
 	input.StartSec = startSec
+	if subtitleStream >= 0 {
+		input.SubtitleStream = &subtitleStream
+	}
 	t.maybeFillDuration(ctx, m, input)
 	if _, err := t.resolveFFmpegPath(); err != nil {
 		return "", err
@@ -177,12 +189,12 @@ func (t *TranscoderService) EnsureJobFrom(ctx context.Context, mediaID string, s
 
 	t.mu.Lock()
 	if existing, ok := t.jobs[mediaID]; ok {
-		if sameHLSStart(existing.startSec, startSec) {
+		if sameHLSConfiguration(existing, startSec, subtitleStream) {
 			t.touchJobLocked(mediaID)
 			t.mu.Unlock()
 			return t.PlaylistPath(mediaID), nil
 		}
-		if !shouldReplaceHLSJob(existing, startSec, seekGen) {
+		if !shouldReplaceHLSJobConfiguration(existing, startSec, seekGen, subtitleStream) {
 			t.touchJobLocked(mediaID)
 			t.mu.Unlock()
 			return t.PlaylistPath(mediaID), nil
@@ -198,16 +210,21 @@ func (t *TranscoderService) EnsureJobFrom(ctx context.Context, mediaID string, s
 	}
 
 	jobCtx, cancel := context.WithCancel(context.Background())
+	jobEncoder := t.effectiveEncoder()
+	if subtitleStream >= 0 {
+		jobEncoder = "software (subtitle burn-in)"
+	}
 	job := &hlsJob{
-		mediaID:    mediaID,
-		outputDir:  outDir,
-		cancel:     cancel,
-		startedAt:  time.Now(),
-		lastAccess: time.Now(),
-		encoder:    t.effectiveEncoder(),
-		startSec:   startSec,
-		seekGen:    seekGen,
-		done:       make(chan struct{}),
+		mediaID:        mediaID,
+		outputDir:      outDir,
+		cancel:         cancel,
+		startedAt:      time.Now(),
+		lastAccess:     time.Now(),
+		encoder:        jobEncoder,
+		startSec:       startSec,
+		seekGen:        seekGen,
+		subtitleStream: subtitleStream,
+		done:           make(chan struct{}),
 	}
 	t.jobs[mediaID] = job
 	t.mu.Unlock()
@@ -229,6 +246,20 @@ func shouldReplaceHLSJob(existing *hlsJob, startSec float64, seekGen int64) bool
 	if sameHLSStart(existing.startSec, startSec) {
 		return false
 	}
+	return newerHLSGenerationMayReplace(existing, seekGen)
+}
+
+func shouldReplaceHLSJobConfiguration(existing *hlsJob, startSec float64, seekGen int64, subtitleStream int) bool {
+	if existing == nil {
+		return true
+	}
+	if sameHLSConfiguration(existing, startSec, subtitleStream) {
+		return false
+	}
+	return newerHLSGenerationMayReplace(existing, seekGen)
+}
+
+func newerHLSGenerationMayReplace(existing *hlsJob, seekGen int64) bool {
 	if seekGen > 0 && existing.seekGen > 0 && seekGen < existing.seekGen {
 		return false
 	}
@@ -239,6 +270,12 @@ func shouldReplaceHLSJob(existing *hlsJob, startSec float64, seekGen int64) bool
 	return true
 }
 
+func sameHLSConfiguration(existing *hlsJob, startSec float64, subtitleStream int) bool {
+	return existing != nil &&
+		existing.subtitleStream == subtitleStream &&
+		sameHLSStart(existing.startSec, startSec)
+}
+
 func (t *TranscoderService) mediaStartGate(mediaID string) *sync.Mutex {
 	v, _ := t.startGates.LoadOrStore(mediaID, &sync.Mutex{})
 	return v.(*sync.Mutex)

internal/service/transcoder_ffmpeg_args.go

@@ -11,9 +11,10 @@ import (
 )
 
 type transcodeInput struct {
-	Source   string
-	Headers  map[string]string
-	StartSec float64
+	Source         string
+	Headers        map[string]string
+	StartSec       float64
+	SubtitleStream *int
 }
 
 type ffmpegArgSettings struct {
@@ -46,6 +47,13 @@ func buildFFmpegArgs(cfg *config.Config, source, playlist, segments string) []st
 func buildFFmpegArgsForInput(cfg *config.Config, input transcodeInput, playlist, segments string) []string {
 	settings := ffmpegArgSettingsFromConfig(cfg)
 	video := ffmpegVideoPlanForSettings(settings)
+	if input.SubtitleStream != nil {
+		// Bitmap subtitles must be composited in software. Keeping CUDA/QSV/
+		// VAAPI frames here would require a download/upload filter chain that
+		// differs by driver and is considerably less portable.
+		settings.encoder = ""
+		video = ffmpegVideoPlanForSettings(settings)
+	}
 
 	// Mid-file restarts must not use -re: output -ss would otherwise crawl to the
 	// seek point at 1x wall-clock before emitting the first HLS segment.
@@ -140,7 +148,16 @@ func appendInputAndVideoArgs(args []string, input transcodeInput, settings ffmpe
 		args = append(args, "-ss", ss)
 	}
 	args = append(args, "-i", input.Source)
-	args = append(args, "-map", "0:v:0?", "-map", "0:a:0?", "-vf", video.filter, "-c:v", video.codec)
+	if input.SubtitleStream != nil {
+		filter := fmt.Sprintf(
+			"[0:v:0][0:%d]overlay=0:0:eof_action=pass,scale=-2:min(%d\\,ih)[v]",
+			*input.SubtitleStream,
+			settings.height,
+		)
+		args = append(args, "-filter_complex", filter, "-map", "[v]", "-map", "0:a:0?", "-c:v", video.codec)
+	} else {
+		args = append(args, "-map", "0:v:0?", "-map", "0:a:0?", "-vf", video.filter, "-c:v", video.codec)
+	}
 	if settings.threads > 0 && video.codec == "libx264" {
 		args = append(args, "-threads", strconv.Itoa(settings.threads))
 	}

internal/service/transcoder_test.go

@@ -255,6 +255,29 @@ func TestBuildFFmpegArgsHTTPSeekAfterDashI(t *testing.T) {
 	}
 }
 
+func TestBuildFFmpegArgsBurnsBitmapSubtitleInSoftware(t *testing.T) {
+	cfg := &config.Config{}
+	cfg.Transcoder.HardwareAccel = true
+	cfg.Transcoder.Encoder = "nvenc"
+	cfg.Transcoder.MaxHeight = 720
+	cfg.Transcoder.SegmentSeconds = 4
+	stream := 3
+	args := buildFFmpegArgsForInput(cfg, transcodeInput{
+		Source:         "/x.mkv",
+		SubtitleStream: &stream,
+	}, "/o/x.m3u8", "/o/seg_%05d.ts")
+	joined := strings.Join(args, " ")
+	if !strings.Contains(joined, "[0:v:0][0:3]overlay=0:0:eof_action=pass") {
+		t.Fatalf("bitmap subtitle overlay missing: %s", joined)
+	}
+	if !strings.Contains(joined, "-map [v]") || !strings.Contains(joined, "-c:v libx264") {
+		t.Fatalf("burn-in should use the filtered software video stream: %s", joined)
+	}
+	if strings.Contains(joined, "cuda") || strings.Contains(joined, "h264_nvenc") {
+		t.Fatalf("burn-in must not retain hardware-only frames: %s", joined)
+	}
+}
+
 func TestSameHLSStart(t *testing.T) {
 	if !sameHLSStart(10, 10.2) {
 		t.Fatal("expected close starts to match")
@@ -283,6 +306,19 @@ func TestShouldReplaceHLSJob(t *testing.T) {
 	}
 }
 
+func TestBitmapSubtitleChangeReplacesHLSJob(t *testing.T) {
+	existing := &hlsJob{startSec: 120, seekGen: 1000, subtitleStream: 2}
+	if !shouldReplaceHLSJobConfiguration(existing, 120, 1001, 3) {
+		t.Fatal("changing bitmap subtitle must replace the HLS generation")
+	}
+	if !shouldReplaceHLSJobConfiguration(existing, 120, 1001, -1) {
+		t.Fatal("closing bitmap subtitle must replace the HLS generation")
+	}
+	if shouldReplaceHLSJobConfiguration(existing, 120.2, 1001, 2) {
+		t.Fatal("same subtitle and nearby start should reuse the HLS generation")
+	}
+}
+
 func TestFilterHLSSegmentQueryDropsStart(t *testing.T) {
 	got := filterHLSSegmentQuery("token=abc&start=120.5&_seek=1001&profile_id=1")
 	if strings.Contains(got, "start=") {

web/src/api/client.ts

@@ -147,7 +147,7 @@ export function streamURL(mediaId: string): string {
 
 // hlsURL returns the m3u8 playlist URL fed into hls.js.
 // startSec > 0 asks the server to (re)start ffmpeg from that source offset.
-export function hlsURL(mediaId: string, startSec = 0): string {
+export function hlsURL(mediaId: string, startSec = 0, subtitleStream?: number): string {
   const safeStart = Math.max(0, Math.round(startSec * 1000) / 1000)
   // Always send start= (including 0) so the server can tell an intentional
   // restart-from-head apart from a missing query on a stale refresh.
@@ -155,7 +155,11 @@ export function hlsURL(mediaId: string, startSec = 0): string {
   // Monotonic-ish client generation: newer seeks win; older in-flight playlist
   // requests must not cancel the active ffmpeg job back to t=0.
   const bust = `&_seek=${Date.now()}`
-  return `/api/hls/${encodeURIComponent(mediaId)}/index.m3u8?${tokenQuery()}${profileQuery()}${start}${bust}`
+  const subtitle =
+    subtitleStream !== undefined && subtitleStream >= 0
+      ? `&subtitle=${encodeURIComponent(String(subtitleStream))}`
+      : ''
+  return `/api/hls/${encodeURIComponent(mediaId)}/index.m3u8?${tokenQuery()}${profileQuery()}${start}${subtitle}${bust}`
 }
 
 // Stop an on-demand HLS job. keepalive makes the request survive page

web/src/api/subtitles.ts

@@ -7,6 +7,9 @@ export interface SubtitleTrack {
   path: string
   url: string
   codec: string
+  source: 'external' | 'embedded'
+  delivery: 'webvtt' | 'burn'
+  stream_index?: number
 }
 
 export const subtitlesAPI = {

web/src/components/PlayerControls.tsx

@@ -30,6 +30,8 @@ function formatTime(s: number): string {
 
 type PlayerControlsProps = {
   videoRef: React.RefObject<HTMLVideoElement>
+  uiVisible: boolean
+  onUiVisibleChange: (visible: boolean) => void
   subs: SubtitleTrack[]
   /** 当前激活字幕轨道：-1=关闭，0..n-1=对应轨道。 */
   subtitleIndex: number
@@ -56,6 +58,8 @@ type PlayerControlsProps = {
 
 export function PlayerControls({
   videoRef,
+  uiVisible,
+  onUiVisibleChange,
   subs,
   subtitleIndex,
   onSelectSubtitle,
@@ -88,7 +92,6 @@ export function PlayerControls({
   const [muted, setMuted] = useState(false)
   const [fullscreen, setFullscreen] = useState(false)
   const [pip, setPip] = useState(false)
-  const [uiVisible, setUiVisible] = useState(true)
   const [controlsHovered, setControlsHovered] = useState(false)
   const [isScrubbing, setIsScrubbing] = useState(false)
   const [scrubValue, setScrubValue] = useState<number | null>(null)
@@ -159,14 +162,14 @@ export function PlayerControls({
             !danmakuOpenRef.current &&
             !playlistOpenRef.current
           ) {
-            setUiVisible(false)
+            onUiVisibleChange(false)
           }
         }, 3000)
       }
     }
 
     const onMove = () => {
-      setUiVisible(true)
+      onUiVisibleChange(true)
       resetTimer()
     }
 
@@ -177,7 +180,7 @@ export function PlayerControls({
       }
       if (el.paused || controlsHoveredRef.current || isScrubbingRef.current || playlistOpenRef.current) return
       if (hideTimerRef.current) clearTimeout(hideTimerRef.current)
-      setUiVisible(false)
+      onUiVisibleChange(false)
     }
 
     const syncPlay = () => {
@@ -248,7 +251,7 @@ export function PlayerControls({
       if (hideTimerRef.current) clearTimeout(hideTimerRef.current)
     }
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [videoRef, knownDuration, streamOffset])
+  }, [videoRef, knownDuration, streamOffset, onUiVisibleChange])
 
   // Keep the scrubber max in sync when metadata duration arrives after mount.
   useEffect(() => {
@@ -260,16 +263,16 @@ export function PlayerControls({
   // 当悬停或菜单状态改变时，更新控制栏计时器
   useEffect(() => {
     if (controlsHovered || isScrubbing || subtitleMenuOpen || danmakuOpen || playlistOpen) {
-      setUiVisible(true)
+      onUiVisibleChange(true)
       if (hideTimerRef.current) clearTimeout(hideTimerRef.current)
     } else {
       const el = video()
       if (el && !el.paused) {
         if (hideTimerRef.current) clearTimeout(hideTimerRef.current)
-        hideTimerRef.current = setTimeout(() => setUiVisible(false), 3000)
+        hideTimerRef.current = setTimeout(() => onUiVisibleChange(false), 3000)
       }
     }
-  }, [controlsHovered, isScrubbing, subtitleMenuOpen, danmakuOpen, playlistOpen])
+  }, [controlsHovered, isScrubbing, subtitleMenuOpen, danmakuOpen, playlistOpen, onUiVisibleChange])
 
   const togglePlay = () => {
     const el = video()
@@ -301,7 +304,7 @@ export function PlayerControls({
 
   const handleSeekStart = () => {
     setIsScrubbing(true)
-    setUiVisible(true)
+    onUiVisibleChange(true)
     if (hideTimerRef.current) clearTimeout(hideTimerRef.current)
   }
 
@@ -361,6 +364,7 @@ export function PlayerControls({
       onMouseEnter={() => setControlsHovered(true)}
       onMouseLeave={() => setControlsHovered(false)}
       onClick={(e) => e.stopPropagation()}
+      onPointerUp={(e) => e.stopPropagation()}
     >
       <div className="flex flex-wrap items-center gap-2 text-white sm:flex-nowrap sm:gap-2.5">
         {/* 上一集 */}

web/src/pages/PlayerPage.tsx

@@ -33,23 +33,6 @@ import { mediaVersionsOf } from '../utils/mediaVersion'
 //
 // External subtitles next to the source file are auto-discovered and
 // attached as <track> elements.
-const SUBTITLE_STORAGE_KEY = 'mebox.subtitle'
-
-// 初始字幕偏好：localStorage 记录上次选择的轨道（-1=关闭）；没有偏好时
-// 默认 0（自动加载第一条字幕）。
-function initialSubtitleIndex(): number {
-  try {
-    const saved = localStorage.getItem(SUBTITLE_STORAGE_KEY)
-    if (saved !== null && saved !== '') {
-      const n = parseInt(saved, 10)
-      if (Number.isFinite(n)) return n
-    }
-  } catch {
-    // ignore
-  }
-  return 0
-}
-
 export function PlayerPage() {
   const { id = '' } = useParams()
   const [params, setParams] = useSearchParams()
@@ -63,13 +46,14 @@ export function PlayerPage() {
   const [media, setMedia] = useState<Media | null>(null)
   const [mode, setMode] = useState<PlayerMode>('direct')
   const [subs, setSubs] = useState<SubtitleTrack[]>([])
-  const [subtitleIndex, setSubtitleIndex] = useState<number>(initialSubtitleIndex)
+  const [subtitleIndex, setSubtitleIndex] = useState<number>(0)
   const [hlsUnavailable, setHlsUnavailable] = useState(false)
   const [playerError, setPlayerError] = useState('')
   // 媒体元数据加载失败（404 / 无权限等）：舞台区直接展示错误而不是永远「加载中」
   const [loadError, setLoadError] = useState('')
   // 「客户端直连解码」模式：宿主机不转码，播放器强制 direct play、隐藏 HLS 切换。
   const [directOnly, setDirectOnly] = useState(false)
+  const [directOnlyKnown, setDirectOnlyKnown] = useState(false)
   const [resumePosition, setResumePosition] = useState(0)
   const [initialSeekDone, setInitialSeekDone] = useState(false)
   // HLS session source offset: playlist t=0 maps to this absolute second.
@@ -121,6 +105,7 @@ export function PlayerPage() {
       .info()
       .then((info) => setDirectOnly(Boolean(info.direct_play_only)))
       .catch(() => setDirectOnly(false))
+      .finally(() => setDirectOnlyKnown(true))
   }, [])
 
   // 读取宿主机已保存的弹幕参数作为面板初始值（无 admin 权限也可读）。
@@ -225,9 +210,8 @@ export function PlayerPage() {
         if (cancelled) return
         const list = tracks ?? []
         setSubs(list)
-        // 记忆的轨道下标可能超出当前媒体的轨道数（不同媒体字幕数量不同），
-        // 越界时回退到第一条；无字幕则关闭。
-        setSubtitleIndex((cur) => (cur >= list.length ? (list.length > 0 ? 0 : -1) : cur))
+        // 服务端始终把外挂字幕排在内嵌字幕前面，因此第一条就是默认优先轨。
+        setSubtitleIndex(list.length > 0 ? 0 : -1)
       })
       .catch(() => {
         if (cancelled) return
@@ -242,6 +226,9 @@ export function PlayerPage() {
   // Depend on media.id (not the media object): refreshing duration after
   // MANIFEST_PARSED must not remount HLS or it storms EnsureJob / DELETE.
   const mediaId = media?.id
+  const selectedSubtitle = subtitleIndex >= 0 ? subs[subtitleIndex] : undefined
+  const burnedSubtitleStream =
+    selectedSubtitle?.delivery === 'burn' ? selectedSubtitle.stream_index : undefined
   const mediaRef = useRef(media)
   mediaRef.current = media
   useEffect(() => {
@@ -254,7 +241,7 @@ export function PlayerPage() {
     const video = ref.current
     const durationSec = currentMedia.duration_sec || 0
     if (mode === 'hls') {
-      const url = hlsURL(mediaId, hlsStartSec)
+      const url = hlsURL(mediaId, hlsStartSec, burnedSubtitleStream)
       void import('hls.js').then(({ default: HlsCtor }) => {
         if (cancelled || !ref.current) return
         if (HlsCtor.isSupported()) {
@@ -323,7 +310,7 @@ export function PlayerPage() {
       cancelled = true
       teardownHls()
     }
-  }, [hlsUnavailable, hlsStartSec, mediaId, mode, params, setParams, teardownHls])
+  }, [burnedSubtitleStream, hlsUnavailable, hlsStartSec, mediaId, mode, params, setParams, teardownHls])
 
   // Stop host ffmpeg when leaving this HLS player. The keepalive request also
   // survives route navigation while the component is being torn down.
@@ -600,6 +587,31 @@ export function PlayerPage() {
 
   const isDirectStream = isDirectStreamMedia(media)
 
+  // 没有外挂字幕且第一条内嵌字幕是图片时，默认轨需要通过 HLS 烧录。
+  useEffect(() => {
+    if (
+      !directOnlyKnown ||
+      directOnly ||
+      isDirectStream ||
+      selectedSubtitle?.delivery !== 'burn' ||
+      mode === 'hls'
+    ) {
+      return
+    }
+    setHlsStartSec(ref.current?.currentTime || 0)
+    setMode('hls')
+    params.set('mode', 'hls')
+    setParams(params, { replace: true })
+  }, [
+    directOnly,
+    directOnlyKnown,
+    isDirectStream,
+    mode,
+    params,
+    selectedSubtitle?.delivery,
+    setParams,
+  ])
+
   const toggleMode = useCallback(() => {
     if (isDirectStream) {
       toast('该媒体为直连播放，无需且不支持转码')
@@ -646,15 +658,27 @@ export function PlayerPage() {
     [hlsStartSec, mode],
   )
 
-  // 用户切换字幕轨道：-1=关闭；记忆偏好，下次播放默认沿用。
+  // 用户切换图片字幕时从当前位置创建新的 HLS 烧录任务；文本字幕只在网页层切换。
   const selectSubtitle = useCallback((index: number) => {
-    setSubtitleIndex(index)
-    try {
-      localStorage.setItem(SUBTITLE_STORAGE_KEY, String(index))
-    } catch {
-      // ignore
+    const oldTrack = subtitleIndex >= 0 ? subs[subtitleIndex] : undefined
+    const nextTrack = index >= 0 ? subs[index] : undefined
+    if (nextTrack?.delivery === 'burn' && (directOnly || isDirectStream)) {
+      toast.error('图片字幕需要开启 HLS 转码后才能显示')
+      return
     }
-  }, [])
+    const burnChanged =
+      oldTrack?.delivery === 'burn' || nextTrack?.delivery === 'burn'
+    if (burnChanged && mode === 'hls' && ref.current) {
+      setHlsStartSec(hlsStartSec + (ref.current.currentTime || 0))
+    }
+    setSubtitleIndex(index)
+    if (nextTrack?.delivery === 'burn' && mode !== 'hls' && !directOnly && !isDirectStream) {
+      setHlsStartSec(ref.current?.currentTime || 0)
+      setMode('hls')
+      params.set('mode', 'hls')
+      setParams(params, { replace: true })
+    }
+  }, [directOnly, hlsStartSec, isDirectStream, mode, params, setParams, subs, subtitleIndex])
 
   const handleVideoError = useCallback(() => {
     // 浏览器对 <video src> 的错误描述非常有限，把详细原因

web/src/pages/PlayerVideoStage.tsx

@@ -1,5 +1,5 @@
 import { useEffect, useRef, useState } from 'react'
-import type { ReactNode, RefObject } from 'react'
+import type { PointerEvent, ReactNode, RefObject } from 'react'
 
 import { subtitlesAPI, type SubtitleTrack } from '../api/subtitles'
 import { type DanmakuAnime, type DanmakuLoadedInfo } from '../api/danmaku'
@@ -85,6 +85,8 @@ export function PlayerVideoStage({
   const stageRef = useRef<HTMLDivElement>(null)
   const [videoRatio, setVideoRatio] = useState<number | null>(null)
   const [stageRect, setStageRect] = useState<{ width: number; height: number } | null>(null)
+  const [controlsVisible, setControlsVisible] = useState(true)
+  const revealControlsOnlyRef = useRef(false)
   // 当前展示的字幕文本（由自定义字幕层渲染，100% 透明无黑框）
   const [activeCueText, setActiveCueText] = useState<string>('')
 
@@ -123,13 +125,26 @@ export function PlayerVideoStage({
     }
   }, [videoRef, media])
 
-  // 点击视频切换播放/暂停；双击切换全屏（控制栏事件自行阻止冒泡）。
+  // 桌面端点击直接切换播放；移动端控制栏隐藏时首次轻触只唤出控制栏，
+  // 控制栏已显示时再次轻触才切换播放/暂停。
   const togglePlay = () => {
     const video = videoRef.current
     if (!video) return
     if (video.paused) void video.play()?.catch(() => undefined)
     else video.pause()
   }
+  const handleStagePointerDown = (event: PointerEvent<HTMLDivElement>) => {
+    revealControlsOnlyRef.current = event.pointerType === 'touch' && !controlsVisible
+  }
+  const handleStageClick = () => {
+    if (revealControlsOnlyRef.current) {
+      revealControlsOnlyRef.current = false
+      setControlsVisible(true)
+      return
+    }
+    revealControlsOnlyRef.current = false
+    togglePlay()
+  }
   const toggleFullscreen = () => {
     const stage = stageRef.current
     if (!stage) return
@@ -142,15 +157,22 @@ export function PlayerVideoStage({
   // 由下方的 React 自定义层输出 100% 纯透明背景、高清晰文字阴影的字幕。
   useEffect(() => {
     const video = videoRef.current
-    if (!video || subs.length === 0 || subtitleIndex < 0 || !subs[subtitleIndex]) {
+    const selectedTrack = subs[subtitleIndex]
+    if (
+      !video ||
+      subs.length === 0 ||
+      subtitleIndex < 0 ||
+      !selectedTrack ||
+      selectedTrack.delivery === 'burn'
+    ) {
       setActiveCueText('')
       return
     }
-    const trackIdx = subtitleIndex
 
     const updateCue = () => {
-      const trackEls = Array.from(video.querySelectorAll('track'))
-      const selectedEl = trackEls[trackIdx]
+      const selectedEl = video.querySelector<HTMLTrackElement>(
+        `track[data-subtitle-index="${subtitleIndex}"]`,
+      )
       const tt = selectedEl?.track
       if (!tt) {
         setActiveCueText('')
@@ -160,13 +182,13 @@ export function PlayerVideoStage({
       // 优先从浏览器 activeCues 中取当前文本；若浏览器在 hidden 模式下延迟触发 cuechange，
       // 则从 tt.cues 中根据 video.currentTime 实时匹配当前字幕，确保初次加载无感立即可见。
       const texts: string[] = []
-      if (tt.activeCues && tt.activeCues.length > 0) {
+      if ((!streamOffset || streamOffset <= 0.05) && tt.activeCues && tt.activeCues.length > 0) {
         for (let i = 0; i < tt.activeCues.length; i++) {
           const cue = tt.activeCues[i] as VTTCue
           if (cue && cue.text) texts.push(cue.text)
         }
       } else if (tt.cues && tt.cues.length > 0) {
-        const cur = video.currentTime
+        const cur = video.currentTime + (streamOffset ?? 0)
         for (let i = 0; i < tt.cues.length; i++) {
           const cue = tt.cues[i] as VTTCue
           if (cue && cur >= cue.startTime && cur <= cue.endTime && cue.text) {
@@ -180,15 +202,18 @@ export function PlayerVideoStage({
     const apply = () => {
       const trackEls = Array.from(video.querySelectorAll('track'))
       if (trackEls.length === 0) return
-      trackEls.forEach((el, i) => {
+      trackEls.forEach((el) => {
         const tt = el.track
         if (tt) {
           // 'hidden' 模式：浏览器解析 WebVTT 并触发 cuechange，但隐藏原生黑底 UI
-          tt.mode = i === trackIdx ? 'hidden' : 'disabled'
+          tt.mode =
+            el.dataset.subtitleIndex === String(subtitleIndex) ? 'hidden' : 'disabled'
         }
       })
 
-      const selected = trackEls[trackIdx]
+      const selected = video.querySelector<HTMLTrackElement>(
+        `track[data-subtitle-index="${subtitleIndex}"]`,
+      )
       if (!selected) return
 
       const tt = selected.track
@@ -216,8 +241,9 @@ export function PlayerVideoStage({
       video.removeEventListener('seeking', updateCue)
       video.removeEventListener('seeked', updateCue)
       video.removeEventListener('playing', updateCue)
-      const trackEls = Array.from(video.querySelectorAll('track'))
-      const selected = trackEls[trackIdx]
+      const selected = video.querySelector<HTMLTrackElement>(
+        `track[data-subtitle-index="${subtitleIndex}"]`,
+      )
       if (selected) {
         selected.removeEventListener('load', updateCue)
         if (selected.track) {
@@ -225,7 +251,7 @@ export function PlayerVideoStage({
         }
       }
     }
-  }, [subtitleIndex, subs, videoRef, media])
+  }, [subtitleIndex, subs, videoRef, media, streamOffset])
 
   // 根据视频画面宽高比与舞台宽高比，确定视频在哪个轴向撑满 100%
   const isWiderThanStage =
@@ -251,7 +277,8 @@ export function PlayerVideoStage({
       ref={stageRef}
       data-player-stage
       className="relative flex h-full w-full flex-1 items-center justify-center overflow-hidden bg-black"
-      onClick={togglePlay}
+      onPointerDown={handleStagePointerDown}
+      onClick={handleStageClick}
       onDoubleClick={toggleFullscreen}
     >
       {media ? (
@@ -267,16 +294,19 @@ export function PlayerVideoStage({
               className="h-full w-full object-contain bg-black"
               onError={onVideoError}
             >
-              {subs.map((track, index) => (
+              {subs.map((track, index) =>
+                track.delivery === 'burn' ? null : (
                 <track
                   key={track.path}
+                  data-subtitle-index={index}
                   kind="subtitles"
                   src={subtitlesAPI.url(media.id, track.path)}
                   srcLang={track.lang}
                   label={track.label || track.lang}
                   default={subtitleIndex === index}
                 />
-              ))}
+                ),
+              )}
             </video>
             <DanmakuStage
               key={media.id}
@@ -310,6 +340,8 @@ export function PlayerVideoStage({
           </div>
           <PlayerControls
             videoRef={videoRef}
+            uiVisible={controlsVisible}
+            onUiVisibleChange={setControlsVisible}
             subs={subs}
             subtitleIndex={subtitleIndex}
             onSelectSubtitle={onSelectSubtitle}