vibedit/export.go

package main

import (
	"context"
	"fmt"
	"log"
	"math"
	"net/http"
	"os"
	"path/filepath"
	"runtime"
	"slices"
	"sort"
	"strconv"
	"strings"
)

type ExportClip struct {
	Src string  `json:"src"`
	In  float64 `json:"in"`
	Out float64 `json:"out"`
}

type ExportRequest struct {
	Clips     []ExportClip `json:"clips"`
	Mode      string       `json:"mode"` // copy | encode | gif
	Join      bool         `json:"join"` // one file, or one file per clip
	Output    string       `json:"output"`
	Overwrite bool         `json:"overwrite"`
	NoAudio   bool         `json:"noAudio"`
	// stream copy
	AllAudio bool `json:"allAudio"`
	// re-encode
	VCodec   string `json:"vcodec"`
	Quality  int    `json:"quality"`
	Speed    string `json:"speed"`  // fast | balanced | slow
	Height   int    `json:"height"` // 0 = keep
	FPS      string `json:"fps"`    // "" = keep
	ACodec   string `json:"acodec"`
	ABitrate int    `json:"abitrate"` // kbps
	// animated GIF (also uses Height and FPS)
	Dither string `json:"dither"` // one of gifDithers
	Loop   bool   `json:"loop"`
}

// gifDithers are the paletteuse dither modes offered for GIF export.
var gifDithers = []string{"sierra2_4a", "bayer", "none"}

type clipJob struct {
	m       *Media
	in, out float64
}

func (c clipJob) dur() float64 { return c.out - c.in }

func (a *App) handleExport(w http.ResponseWriter, r *http.Request) {
	if !a.tools.OK() {
		writeError(w, http.StatusServiceUnavailable, "ffmpeg/ffprobe not found")
		return
	}
	var req ExportRequest
	if err := readJSON(r, &req); err != nil {
		writeError(w, http.StatusBadRequest, "invalid request: "+err.Error())
		return
	}
	if a.jobs.AnyRunning("export") {
		writeError(w, http.StatusConflict, "an export is already running")
		return
	}
	clips, err := a.resolveClips(req.Clips)
	if err != nil {
		writeError(w, http.StatusBadRequest, err.Error())
		return
	}
	out, err := filepath.Abs(strings.TrimSpace(req.Output))
	if err != nil || strings.TrimSpace(req.Output) == "" {
		writeError(w, http.StatusBadRequest, "choose where to save the export")
		return
	}
	ext := strings.ToLower(strings.TrimPrefix(filepath.Ext(out), "."))
	if st, err := os.Stat(filepath.Dir(out)); err != nil || !st.IsDir() {
		writeError(w, http.StatusBadRequest, "the output folder does not exist: "+filepath.Dir(out))
		return
	}

	switch req.Mode {
	case "copy":
		if !slices.Contains(copyExts, ext) {
			writeError(w, http.StatusBadRequest, fmt.Sprintf("unsupported output type %q for lossless export", "."+ext))
			return
		}
		if req.Join && len(clips) > 1 {
			if msg := copyCompatProblem(clips, req.AllAudio && !req.NoAudio); msg != "" {
				writeError(w, http.StatusBadRequest, msg)
				return
			}
		}
	case "encode":
		if err := a.validateEncode(&req, ext); err != nil {
			writeError(w, http.StatusBadRequest, err.Error())
			return
		}
	case "gif":
		if err := validateGIF(&req, ext); err != nil {
			writeError(w, http.StatusBadRequest, err.Error())
			return
		}
	default:
		writeError(w, http.StatusBadRequest, "unknown export mode")
		return
	}

	outputs := []string{out}
	if !req.Join {
		outputs = numberedOutputs(out, len(clips))
	}
	for _, o := range outputs {
		for _, c := range clips {
			if samePath(o, c.m.Path) {
				writeError(w, http.StatusBadRequest, "the output would overwrite a source video: "+o)
				return
			}
		}
	}
	if !req.Overwrite {
		var exists []string
		for _, o := range outputs {
			if _, err := os.Stat(o); err == nil {
				exists = append(exists, o)
			}
		}
		if len(exists) > 0 {
			writeErrorData(w, http.StatusConflict, "output file already exists", map[string]any{"exists": exists})
			return
		}
	}

	title := filepath.Base(out)
	if !req.Join {
		title = fmt.Sprintf("%d files", len(outputs))
	}
	log.Printf("export started: %d clip(s), mode=%s, join=%v -> %s", len(clips), req.Mode, req.Join, out)
	j := a.jobs.Start("export", title, "", func(ctx context.Context, c *JobCtl) error {
		return a.runExport(ctx, c, &req, clips, outputs)
	})
	writeJSON(w, map[string]any{"jobId": j.ID, "outputs": outputs})
}

func (a *App) resolveClips(in []ExportClip) ([]clipJob, error) {
	if len(in) == 0 {
		return nil, fmt.Errorf("there are no clips to export")
	}
	var out []clipJob
	for i, c := range in {
		m := a.media.Get(c.Src)
		if m == nil {
			return nil, fmt.Errorf("clip %d: source video is not loaded", i+1)
		}
		cin := math.Max(0, c.In)
		cout := math.Min(c.Out, m.Duration)
		if cout-cin < 0.01 {
			return nil, fmt.Errorf("clip %d is empty (in %.3f, out %.3f)", i+1, c.In, c.Out)
		}
		out = append(out, clipJob{m: m, in: cin, out: cout})
	}
	return out, nil
}

// numberedOutputs turns "C:\v\name.mp4" into name_01.mp4, name_02.mp4, ...
func numberedOutputs(out string, n int) []string {
	ext := filepath.Ext(out)
	base := strings.TrimSuffix(out, ext)
	width := len(strconv.Itoa(n))
	if width < 2 {
		width = 2
	}
	paths := make([]string, n)
	for i := range paths {
		paths[i] = fmt.Sprintf("%s_%0*d%s", base, width, i+1, ext)
	}
	return paths
}

func samePath(a, b string) bool {
	a, b = filepath.Clean(a), filepath.Clean(b)
	if runtime.GOOS == "windows" || runtime.GOOS == "darwin" {
		return strings.EqualFold(a, b)
	}
	return a == b
}

// copyCompatProblem explains why clips from different sources cannot be
// joined without re-encoding, or returns "".
func copyCompatProblem(clips []clipJob, allAudio bool) string {
	sig := func(m *Media) string {
		var parts []string
		if v := m.Video; v != nil {
			parts = append(parts, fmt.Sprintf("video %s %s %dx%d %s", v.Codec, v.Profile, v.Width, v.Height, v.PixFmt))
		} else {
			parts = append(parts, "no video")
		}
		audio := m.Audio
		if !allAudio && len(audio) > 1 {
			audio = audio[:1]
		}
		for _, s := range audio {
			parts = append(parts, fmt.Sprintf("audio %s %dHz %dch", s.Codec, s.SampleRate, s.Channels))
		}
		if len(audio) == 0 {
			parts = append(parts, "no audio")
		}
		return strings.Join(parts, ", ")
	}
	first := clips[0].m
	want := sig(first)
	for _, c := range clips[1:] {
		if c.m == first {
			continue
		}
		if got := sig(c.m); got != want {
			return fmt.Sprintf("Lossless join needs all sources to have the same format, but they differ:\n• %s: %s\n• %s: %s\nUse Re-encode, or export each clip separately.",
				first.Name, want, c.m.Name, got)
		}
	}
	return ""
}

func (a *App) validateEncode(req *ExportRequest, ext string) error {
	ct, ok := containerByExt(ext)
	if !ok {
		return fmt.Errorf("re-encoding supports .mp4, .mkv, .mov and .webm outputs, not %q", "."+ext)
	}
	vc, ok := codecByName(req.VCodec)
	if !ok || vc.Kind != "video" || !a.tools.HasEncoder(vc.Name) {
		return fmt.Errorf("video encoder %q is not available", req.VCodec)
	}
	if !slices.Contains(ct.Video, vc.Family) {
		return fmt.Errorf("%s cannot contain %s video", ct.Label, vc.Label)
	}
	if req.ACodec == "" || req.ACodec == "none" {
		req.NoAudio = true
	}
	if !req.NoAudio {
		ac, ok := codecByName(req.ACodec)
		if !ok || ac.Kind != "audio" || !a.tools.HasEncoder(ac.Name) {
			return fmt.Errorf("audio encoder %q is not available", req.ACodec)
		}
		if !slices.Contains(ct.Audio, ac.Family) {
			return fmt.Errorf("%s cannot contain %s audio", ct.Label, ac.Label)
		}
	}
	if req.Quality < 0 || req.Quality > vc.MaxQ {
		req.Quality = vc.DefQ
	}
	if req.Height < 0 || req.Height > 8640 {
		return fmt.Errorf("invalid output height %d", req.Height)
	}
	if req.FPS != "" && parseRatio(req.FPS, "/") <= 0 {
		return fmt.Errorf("invalid frame rate %q", req.FPS)
	}
	if req.ABitrate <= 0 || req.ABitrate > 1024 {
		req.ABitrate = 192
	}
	return nil
}

func validateGIF(req *ExportRequest, ext string) error {
	if ext != "gif" {
		return fmt.Errorf("animated GIF export needs a .gif output, not %q", "."+ext)
	}
	if req.Height < 0 || req.Height > 8640 {
		return fmt.Errorf("invalid output height %d", req.Height)
	}
	// GIF frame delays are whole centiseconds, and browsers slow down
	// anything faster than 50 fps.
	if fps := parseRatio(req.FPS, "/"); fps <= 0 || fps > 50 {
		return fmt.Errorf("invalid GIF frame rate %q (up to 50 fps)", req.FPS)
	}
	if !slices.Contains(gifDithers, req.Dither) {
		req.Dither = gifDithers[0]
	}
	req.NoAudio = true
	return nil
}

func (a *App) runExport(ctx context.Context, c *JobCtl, req *ExportRequest, clips []clipJob, outputs []string) error {
	var total float64
	for _, cl := range clips {
		total += cl.dur()
	}
	copyJoin := req.Mode == "copy" && req.Join && len(clips) > 1
	if copyJoin {
		total *= 2 // cutting pass + joining pass
	}
	done := 0.0
	progress := func(base, limit float64) func(float64, string) {
		return func(sec float64, speed string) {
			c.Progress((base+math.Min(sec, limit))/total, speed)
		}
	}
	// run writes to a temporary name and renames on success, so a failed or
	// canceled export never leaves a broken file (or destroys an old one).
	run := func(args []string, out string, dur float64) error {
		tmp := partialName(out)
		args = append(args, tmp)
		err := a.tools.runFFmpeg(ctx, args, progress(done, dur))
		if err != nil {
			os.Remove(tmp)
			return err
		}
		done += dur
		if err := os.Rename(tmp, out); err != nil {
			os.Remove(tmp)
			return err
		}
		return nil
	}

	keyframes := map[*Media][]float64{}
	if req.Mode == "copy" {
		c.Step("Reading keyframes")
		for _, cl := range clips {
			if _, ok := keyframes[cl.m]; ok {
				continue
			}
			kfs, err := a.media.Keyframes(ctx, cl.m)
			if ctx.Err() != nil {
				return ctx.Err()
			}
			if err != nil {
				log.Printf("keyframe scan failed for %s, cutting without it: %v", cl.m.Name, err)
			}
			keyframes[cl.m] = kfs
		}
	}

	switch {
	case copyJoin:
		out := outputs[0]
		tmpDir, err := os.MkdirTemp(filepath.Dir(out), ".vibedit-tmp-")
		if err != nil {
			return fmt.Errorf("could not create a temporary folder next to the output: %w", err)
		}
		defer os.RemoveAll(tmpDir)
		ext := filepath.Ext(out)
		var list strings.Builder
		for i, cl := range clips {
			c.Step(fmt.Sprintf("Cutting clip %d of %d", i+1, len(clips)))
			name := fmt.Sprintf("seg_%04d%s", i+1, ext)
			seg := filepath.Join(tmpDir, name)
			args := append(copyClipArgs(cl, req, keyframes[cl.m]), copyOutputArgs(ext, false)...)
			if err := a.tools.runFFmpeg(ctx, append(args, seg), progress(done, cl.dur())); err != nil {
				return fmt.Errorf("clip %d: %w", i+1, err)
			}
			done += cl.dur()
			fmt.Fprintf(&list, "file '%s'\n", name)
		}
		listPath := filepath.Join(tmpDir, "list.txt")
		if err := os.WriteFile(listPath, []byte(list.String()), 0o644); err != nil {
			return err
		}
		c.Step("Joining clips")
		args := ffmpegBaseArgs()
		args = append(args, "-f", "concat", "-safe", "0", "-i", listPath, "-map", "0", "-c", "copy")
		args = append(args, copyOutputArgs(ext, true)...)
		if err := run(args, out, total/2); err != nil {
			return err
		}
		c.AddOutput(out)

	case req.Mode == "copy":
		for i, cl := range clips {
			c.Step(stepName("Saving", i, len(clips)))
			args := append(copyClipArgs(cl, req, keyframes[cl.m]), copyOutputArgs(filepath.Ext(outputs[i]), true)...)
			if err := run(args, outputs[i], cl.dur()); err != nil {
				return clipErr(err, i, len(clips))
			}
			c.AddOutput(outputs[i])
		}

	case req.Mode == "gif":
		tmpDir, err := os.MkdirTemp("", "vibedit-gif-")
		if err != nil {
			return err
		}
		defer os.RemoveAll(tmpDir)
		palette := filepath.Join(tmpDir, "palette.png")
		groups := [][]clipJob{clips}
		if !req.Join {
			groups = nil
			for _, cl := range clips {
				groups = append(groups, []clipJob{cl})
			}
		}
		for i, g := range groups {
			of := ""
			if len(groups) > 1 {
				of = fmt.Sprintf(" (clip %d of %d)", i+1, len(groups))
			}
			paletteRun, gifRun, err := gifArgs(g, req, palette)
			if err != nil {
				return err
			}
			// ffmpeg reports no progress while it builds a palette, so like
			// the keyframe scan this step does not move the progress bar.
			c.Step("Building the color palette" + of)
			if err := a.tools.runFFmpeg(ctx, append(paletteRun, palette), nil); err != nil {
				return clipErr(err, i, len(groups))
			}
			c.Step("Encoding GIF" + of)
			var dur float64
			for _, cl := range g {
				dur += cl.dur()
			}
			if err := run(gifRun, outputs[i], dur); err != nil {
				return clipErr(err, i, len(groups))
			}
			c.AddOutput(outputs[i])
		}

	case req.Join:
		c.Step("Encoding")
		args, err := encodeArgs(clips, req, filepath.Ext(outputs[0]))
		if err != nil {
			return err
		}
		if err := run(args, outputs[0], total); err != nil {
			return err
		}
		c.AddOutput(outputs[0])

	default:
		for i, cl := range clips {
			c.Step(stepName("Encoding", i, len(clips)))
			args, err := encodeArgs([]clipJob{cl}, req, filepath.Ext(outputs[i]))
			if err != nil {
				return err
			}
			if err := run(args, outputs[i], cl.dur()); err != nil {
				return clipErr(err, i, len(clips))
			}
			c.AddOutput(outputs[i])
		}
	}
	return nil
}

func stepName(verb string, i, n int) string {
	if n == 1 {
		return verb
	}
	return fmt.Sprintf("%s clip %d of %d", verb, i+1, n)
}

func clipErr(err error, i, n int) error {
	if n == 1 || err == context.Canceled {
		return err
	}
	return fmt.Errorf("clip %d: %w", i+1, err)
}

// partialName keeps the extension last so ffmpeg still picks the right muxer.
func partialName(out string) string {
	ext := filepath.Ext(out)
	return strings.TrimSuffix(out, ext) + ".vibedit-part" + ext
}

// copyClipArgs cuts one clip without re-encoding. Stream copy can only start
// on a keyframe, so the clip starts at the keyframe at or before In. When the
// keyframe positions are known we seek exactly to it and drop packets before
// it, so audio does not start ahead of the video.
func copyClipArgs(cl clipJob, req *ExportRequest, keyframes []float64) []string {
	start := cl.in
	exact := false
	if i := sort.SearchFloat64s(keyframes, cl.in+0.001); i > 0 {
		start, exact = keyframes[i-1], true
	}
	args := ffmpegBaseArgs()
	if exact && start <= 0.001 {
		start = 0
	} else if exact {
		start -= 0.0005 // stay just before the keyframe despite rounding
		args = append(args, "-ss", ffTime(start))
	} else {
		args = append(args, "-ss", ffTime(start))
	}
	args = append(args, "-t", ffTime(cl.out-start), "-i", cl.m.Path)
	if v := cl.m.Video; v != nil {
		args = append(args, "-map", fmt.Sprintf("0:%d", v.Index))
	}
	if !req.NoAudio {
		for i, s := range cl.m.Audio {
			if i > 0 && !req.AllAudio {
				break
			}
			args = append(args, "-map", fmt.Sprintf("0:%d", s.Index))
		}
	}
	args = append(args, "-c", "copy", "-avoid_negative_ts", "make_zero", "-map_metadata", "0", "-map_chapters", "-1")
	if exact {
		args = append(args, "-copypriorss", "0")
	}
	return args
}

func copyOutputArgs(ext string, final bool) []string {
	ext = strings.ToLower(ext)
	if final && (ext == ".mp4" || ext == ".mov" || ext == ".m4v") {
		return []string{"-movflags", "+faststart"}
	}
	return nil
}

// encodeArgs builds one ffmpeg command that decodes every clip (each as its
// own seeked input), normalizes size/frame rate/audio format and joins them
// with the concat filter.
func encodeArgs(clips []clipJob, req *ExportRequest, ext string) ([]string, error) {
	f, err := frameFormat(clips, req.Height, req.FPS)
	if err != nil {
		return nil, err
	}
	anyAudio := false
	for _, cl := range clips {
		if len(cl.m.Audio) > 0 {
			anyAudio = true
		}
	}
	anyAudio = anyAudio && !req.NoAudio

	vc, _ := codecByName(req.VCodec)
	var venc []string
	venc, f.PixFmt = videoEncoderArgs(vc, req.Quality, req.Speed)
	if anyAudio {
		f.SampleRate, f.Layout = audioFormat(clips, req.ACodec)
	}

	inputs, graph, _ := joinGraph(clips, f)
	args := append(ffmpegBaseArgs(), inputs...)
	args = append(args, "-filter_complex", graph, "-map", "[outv]")
	if anyAudio {
		args = append(args, "-map", "[outa]")
	}
	args = append(args, venc...)
	ext = strings.ToLower(ext)
	if vc.Family == "hevc" && (ext == ".mp4" || ext == ".mov") {
		args = append(args, "-tag:v", "hvc1") // plays on Apple devices too
	}
	if anyAudio {
		args = append(args, "-c:a", req.ACodec)
		if req.ACodec != "flac" {
			args = append(args, "-b:a", fmt.Sprintf("%dk", req.ABitrate))
		}
	}
	args = append(args, "-map_chapters", "-1")
	if ext == ".mp4" || ext == ".mov" {
		args = append(args, "-movflags", "+faststart")
	}
	return args, nil
}

// gifArgs builds the two ffmpeg runs of a GIF export, each still missing its
// output path. The first picks the 256 colors that suit the whole animation
// best and writes them to palette; the second maps every frame onto them.
func gifArgs(clips []clipJob, req *ExportRequest, palette string) (paletteRun, gifRun []string, err error) {
	f, err := frameFormat(clips, req.Height, req.FPS)
	if err != nil {
		return nil, nil, err
	}
	f.PixFmt = "bgra" // what palettegen and paletteuse take
	inputs, graph, n := joinGraph(clips, f)

	paletteRun = append(ffmpegBaseArgs(), inputs...)
	paletteRun = append(paletteRun, "-filter_complex", graph+";[outv]palettegen[pal]", "-map", "[pal]", "-frames:v", "1", "-update", "1")

	// diff_mode=rectangle re-dithers only the part of a frame that changed,
	// which keeps the dither noise still and the file smaller.
	loop := "0" // forever
	if !req.Loop {
		loop = "-1"
	}
	gifRun = append(ffmpegBaseArgs(), inputs...)
	gifRun = append(gifRun, "-i", palette,
		"-filter_complex", fmt.Sprintf("%s;[outv][%d:v]paletteuse=dither=%s:diff_mode=rectangle[gif]", graph, n, req.Dither),
		"-map", "[gif]", "-loop", loop)
	return paletteRun, gifRun, nil
}

// outFormat is what joinGraph converts every clip to. SampleRate 0 means no
// audio.
type outFormat struct {
	W, H       int
	FPS        string
	PixFmt     string
	SampleRate int
	Layout     string
}

// frameFormat picks the output frame size and rate: the first video clip's
// display size, scaled to height when set, and its frame rate unless fps is
// set.
func frameFormat(clips []clipJob, height int, fps string) (outFormat, error) {
	var first *Media
	for _, cl := range clips {
		if cl.m.Video != nil {
			first = cl.m
			break
		}
	}
	if first == nil {
		return outFormat{}, fmt.Errorf("none of the clips contain video")
	}
	dw, dh := float64(first.Video.DisplayWidth), float64(first.Video.DisplayHeight)
	if dw <= 0 || dh <= 0 {
		dw, dh = 1280, 720
	}
	if height > 0 && float64(height) != dh {
		dw, dh = dw*float64(height)/dh, float64(height)
	}
	if fps == "" {
		fps = first.Video.FrameRate
	}
	return outFormat{W: even(dw), H: even(dh), FPS: fps}, nil
}

// joinGraph returns the input arguments (one seeked input per clip, plus
// generated black video or silence for clips that lack it), how many inputs
// that is, and a filter graph that converts every clip to f, letterboxing as
// needed, and joins them with the concat filter into [outv] (and [outa]).
func joinGraph(clips []clipJob, f outFormat) (args []string, graph string, inputs int) {
	var filters []string
	var concatIn strings.Builder
	for i, cl := range clips {
		// Start half a millisecond early so rounding never drops the In frame;
		// the end (start + duration) then falls just before the Out frame.
		d := ffTime(cl.dur())
		args = append(args, "-ss", ffTime(math.Max(0, cl.in-0.0005)), "-t", d, "-i", cl.m.Path)
		src := inputs
		inputs++

		if v := cl.m.Video; v != nil {
			chain := fmt.Sprintf("[%d:%d]", src, v.Index)
			if !v.SquarePixels {
				chain += "scale='trunc(iw*sar/2)*2':ih,setsar=1,"
			}
			chain += fmt.Sprintf("scale=%d:%d:force_original_aspect_ratio=decrease:force_divisible_by=2,pad=%d:%d:(ow-iw)/2:(oh-ih)/2:color=black,setsar=1,fps=%s,format=%s[v%d]",
				f.W, f.H, f.W, f.H, f.FPS, f.PixFmt, i)
			filters = append(filters, chain)
		} else {
			args = append(args, "-f", "lavfi", "-t", d, "-i", fmt.Sprintf("color=c=black:s=%dx%d:r=%s", f.W, f.H, f.FPS))
			filters = append(filters, fmt.Sprintf("[%d:v]setsar=1,format=%s[v%d]", inputs, f.PixFmt, i))
			inputs++
		}
		fmt.Fprintf(&concatIn, "[v%d]", i)

		if f.SampleRate > 0 {
			aformat := fmt.Sprintf("aresample=%d,aformat=sample_fmts=fltp:sample_rates=%d:channel_layouts=%s", f.SampleRate, f.SampleRate, f.Layout)
			if s := cl.m.FirstAudio(); s != nil {
				filters = append(filters, fmt.Sprintf("[%d:%d]%s[a%d]", src, s.Index, aformat, i))
			} else {
				args = append(args, "-f", "lavfi", "-t", d, "-i", fmt.Sprintf("anullsrc=r=%d:cl=%s", f.SampleRate, f.Layout))
				filters = append(filters, fmt.Sprintf("[%d:a]%s[a%d]", inputs, aformat, i))
				inputs++
			}
			fmt.Fprintf(&concatIn, "[a%d]", i)
		}
	}
	concat := fmt.Sprintf("%sconcat=n=%d:v=1:a=0[outv]", concatIn.String(), len(clips))
	if f.SampleRate > 0 {
		concat = fmt.Sprintf("%sconcat=n=%d:v=1:a=1[outv][outa]", concatIn.String(), len(clips))
	}
	filters = append(filters, concat)
	return args, strings.Join(filters, ";"), inputs
}

func even(f float64) int {
	n := int(math.Round(f))
	if n%2 != 0 {
		n++
	}
	if n < 2 {
		n = 2
	}
	return n
}

// audioFormat picks a common sample rate and channel layout for all clips.
func audioFormat(clips []clipJob, acodec string) (int, string) {
	rate, layout := 0, ""
	sameRate, sameLayout := true, true
	for _, cl := range clips {
		s := cl.m.FirstAudio()
		if s == nil {
			continue
		}
		l := s.ChannelLayout
		if l == "" {
			l = map[int]string{1: "mono", 2: "stereo"}[s.Channels]
		}
		if rate == 0 {
			rate, layout = s.SampleRate, l
			continue
		}
		sameRate = sameRate && s.SampleRate == rate
		sameLayout = sameLayout && l == layout
	}
	if !sameRate || (rate != 44100 && rate != 48000) || acodec == "libopus" {
		rate = 48000
	}
	surround := layout == "5.1" || layout == "5.1(side)" || layout == "7.1"
	switch {
	case sameLayout && layout == "mono":
	case sameLayout && surround && (acodec == "aac" || acodec == "flac"):
	default:
		layout = "stereo"
	}
	return rate, layout
}

// videoEncoderArgs maps the generic quality (CRF-like, lower is better) and
// speed setting onto each encoder's own options.
func videoEncoderArgs(c Codec, q int, speed string) ([]string, string) {
	pick := func(fast, balanced, slow string) string {
		switch speed {
		case "fast":
			return fast
		case "slow":
			return slow
		}
		return balanced
	}
	qs := strconv.Itoa(q)
	args := []string{"-c:v", c.Name}
	pixFmt := "yuv420p"
	switch {
	case c.Name == "libx264" || c.Name == "libx265":
		args = append(args, "-crf", qs, "-preset", pick("veryfast", "medium", "slow"))
		if c.Name == "libx265" {
			args = append(args, "-x265-params", "log-level=error")
		}
	case c.Name == "libsvtav1":
		args = append(args, "-crf", qs, "-preset", pick("10", "7", "4"))
	case c.Name == "libaom-av1":
		args = append(args, "-crf", qs, "-b:v", "0", "-cpu-used", pick("8", "6", "4"), "-row-mt", "1")
	case c.Name == "libvpx-vp9":
		args = append(args, "-crf", qs, "-b:v", "0", "-deadline", "good", "-cpu-used", pick("5", "2", "1"), "-row-mt", "1")
	case strings.HasSuffix(c.Name, "_nvenc"):
		args = append(args, "-rc", "vbr", "-cq", qs, "-b:v", "0", "-preset", pick("p2", "p4", "p6"))
	case strings.HasSuffix(c.Name, "_qsv"):
		args = append(args, "-global_quality", qs, "-preset", pick("veryfast", "medium", "slow"))
		pixFmt = "nv12"
	case strings.HasSuffix(c.Name, "_amf"):
		args = append(args, "-rc", "cqp", "-qp_i", qs, "-qp_p", qs, "-quality", pick("speed", "balanced", "quality"))
		if c.Family == "h264" {
			args = append(args, "-qp_b", qs)
		}
	}
	return args, pixFmt
}