ffmpeg-skill 3.2.2 → 3.3.0

package/scripts/_smartcut.py

@@ -0,0 +1,430 @@
+"""cut.py --smart: a keyframe-aware cut that re-encodes only the partial GOPs at each cut point.
+
+A segment [s, e) on a CFR, progressive H.264 source is split at its keyframes:
+
+    head  [first frame at/after s, K1)        re-encoded (K1 = first keyframe at or after s)
+    copy  [K1, K2)                            stream-copied, whole closed GOPs
+    tail  [K2, first frame at/after e)        re-encoded (K2 = last keyframe at or before e)
+
+A part that is empty (s on a keyframe, e on a keyframe or the end of the video) is left out; a
+segment with no keyframe inside it is one re-encoded part. The re-encoded parts are encoded with
+libx264 matched to the source stream (profile, level, pixel format, reorder depth, frame size,
+SAR, colour tags) and every part is written as MPEG-TS, where each keyframe carries its own
+parameter sets in-band, so the decoder switches between the encoder's SPS/PPS and the source's at
+every join. The parts are joined with the concat demuxer (each part's exact length stated), and
+the sound is cut from the source on the same frame grid and encoded once, in the same command.
+
+The result is verified before it is moved into place, and any failure falls back to the full
+re-encode with the reason stated:
+
+    frames    the output's frame count and presentation times: one frame per frame duration from
+              0, none missing, none repeated
+    packets   every copied part's packets against the source's (MD5 of each packet with the
+              parameter-set and delimiter NAL units removed, which the in-band switch adds)
+    decode    framemd5 of each re-encoded part as written against the same frames decoded from
+              the output, and of the first DECODE_CHECK_MAX_FRAMES frames of each copied
+              part against the source's
+    params    the re-encoded parts' stream parameters against the source's, and the output's
+              time base against the source's
+
+Nothing here decides where to cut; it only executes the cut it is given.
+"""
+import os
+from fractions import Fraction
+from typing import Any, Dict, List, Optional, Tuple
+
+from _common import STATE, aac_args, concat_list_line, ffmpeg_base, require_tool, run
+from _common.runner import fail_if_classified
+
+# the stream parameters a re-encoded part must share with the source (ffprobe -show_streams)
+MATCH_FIELDS = ("codec_name", "profile", "level", "pix_fmt", "width", "height", "sample_aspect_ratio",
+                "has_b_frames", "color_space", "color_range", "color_transfer", "color_primaries")
+# what libx264 here can encode to match: 8-bit 4:2:0 (full range as yuvj420p)
+SMART_PIX_FMTS = ("yuv420p", "yuvj420p")
+SMART_PROFILES = {"Constrained Baseline": "baseline", "Baseline": "baseline", "Main": "main", "High": "high"}
+SMART_EXTS = (".mp4", ".mov", ".m4v", ".mkv")
+# the first frames of a copied part are decoded and compared against the source's; capped so a long part
+# does not make the check cost a long decode
+DECODE_CHECK_MAX_FRAMES = 300
+# an input -ss lands this far after the keyframe it is asked for (ffprobe prints microseconds)
+_SEEK_PAST = 1e-4
+_DTS_MARGIN = 1e-4
+
+
+class SmartFallback(Exception):
+    """The smart cut cannot be made or did not verify: `reason` is the stated cause."""
+
+    def __init__(self, reason: str, code: str):
+        super().__init__(reason)
+        self.reason, self.code = reason, code
+
+
+def stream_params(path: str) -> Optional[Dict[str, Any]]:
+    """The first video stream's parameters (MATCH_FIELDS plus field_order, time_base, r_frame_rate)."""
+    import json
+    fields = ",".join(MATCH_FIELDS + ("field_order", "time_base", "r_frame_rate"))
+    proc = run([require_tool("ffprobe"), "-v", "error", "-select_streams", "v:0", "-show_entries",
+                f"stream={fields}", "-of", "json", path], quiet=True, check=False)
+    try:
+        streams = json.loads(proc.stdout or "{}").get("streams") or []
+    except ValueError:
+        return None
+    return streams[0] if proc.returncode == 0 and streams else None
+
+
+def _norm(field: str, value: Any) -> Any:
+    if field == "profile":
+        return SMART_PROFILES.get(str(value), value)
+    if field == "sample_aspect_ratio" and value in (None, "0:1", "N/A"):
+        return "1:1"
+    return value
+
+
+def mismatches(want: Dict[str, Any], got: Dict[str, Any]) -> List[str]:
+    """`field source/part` for every MATCH_FIELDS value the part does not share."""
+    return [f"{f} {want.get(f)}/{got.get(f)}" for f in MATCH_FIELDS if _norm(f, want.get(f)) != _norm(f, got.get(f))]
+
+
+def ineligible(params: Optional[Dict[str, Any]], meta: Dict[str, Any], ext: str) -> Optional[Tuple[str, str]]:
+    """(reason, code) when the source or the output cannot take a smart cut, else None."""
+    if params is None:
+        return "the source's video stream parameters could not be read", "params_unreadable"
+    if ext.lower() not in SMART_EXTS:
+        return f"--smart writes {', '.join(SMART_EXTS)}; {ext} is not one", "container"
+    if params.get("codec_name") != "h264":
+        return (f"--smart matches the source with libx264, so it takes H.264; this source is "
+                f"{params.get('codec_name')}", "codec")
+    if params.get("pix_fmt") not in SMART_PIX_FMTS:
+        return f"--smart encodes 8-bit 4:2:0; this source is {params.get('pix_fmt')}", "pix_fmt"
+    if str(params.get("profile")) not in SMART_PROFILES:
+        return f"H.264 profile {params.get('profile')} cannot be matched by libx264 here", "profile"
+    if params.get("field_order") not in (None, "progressive", "unknown"):
+        return f"the source is interlaced ({params.get('field_order')})", "interlaced"
+    if int(params.get("has_b_frames") or 0) > 2:
+        return f"a reorder depth of {params.get('has_b_frames')} frames cannot be matched", "reorder_depth"
+    if not _rate(params):
+        return "the source states no frame rate", "frame_rate"
+    if (meta.get("video") or {}).get("bt2020_or_hdr"):
+        return "an HDR or BT.2020 source is encoded on the 10-bit HEVC line, not matched", "hdr"
+    return None
+
+
+def _rate(params: Dict[str, Any]) -> Optional[Fraction]:
+    try:
+        q = Fraction(str(params.get("r_frame_rate") or ""))
+    except (ValueError, ZeroDivisionError):
+        return None
+    return q if q > 0 else None
+
+
+def x264_line(params: Dict[str, Any], crf: int, preset: str) -> List[str]:
+    """libx264 arguments that reproduce the source stream's parameters (MATCH_FIELDS)."""
+    level = int(params.get("level") or 0)
+    depth = int(params.get("has_b_frames") or 0)
+    # the reorder depth decides how far each part's decode times run ahead of its presentation
+    # times; parts of different depth would collide in decode order at the join
+    b = "bframes=0" if depth == 0 else ("bframes=3:b-pyramid=none" if depth == 1 else "bframes=3:b-pyramid=normal")
+    args = ["-c:v", "libx264", "-preset", preset, "-crf", str(crf),
+            "-profile:v", SMART_PROFILES[str(params["profile"])], "-pix_fmt", params["pix_fmt"],
+            "-x264-params", f"{b}:open-gop=0"]
+    if level:
+        args += ["-level:v", f"{level // 10}.{level % 10}"]
+    for flag, field in (("-colorspace", "color_space"), ("-color_primaries", "color_primaries"),
+                        ("-color_trc", "color_transfer"), ("-color_range", "color_range")):
+        value = params.get(field)
+        if value and value != "unknown":
+            args += [flag, str(value)]
+    return args
+
+
+def frames_between(frames: List[float], lo: float, hi: Optional[float]) -> List[float]:
+    return [p for p in frames if p >= lo - 1e-6 and (hi is None or p < hi - 1e-6)]
+
+
+def plan_segment(packets: List[Tuple[Any, Any, bool]], s: float, e: float, vend: Optional[float],
+                 gop_is_open) -> List[Dict[str, Any]]:
+    """The parts of one segment, in order: {kind: reencode|copy, start, end (None: to the end of
+    the video), frames}. Raises SmartFallback when a keyframe the copy would start or stop at
+    opens its GOP or the packets around a cut were not read."""
+    frames = sorted(p[0] for p in packets if p[0] is not None)
+    keyidx = [i for i, p in enumerate(packets) if p[2] and p[0] is not None and p[1] is not None]
+    seg = frames_between(frames, s, e)
+    if not seg:
+        raise SmartFallback(f"no frame of the source lies in {s:.3f}-{e:.3f}s", "no_frames")
+    first, last = seg[0], seg[-1]
+    to_end = vend is not None and e >= vend - 1e-6
+    end_bound = None if to_end else next((p for p in frames if p >= e - 1e-6), None)
+    if end_bound is None and not to_end:
+        raise SmartFallback(f"the frames after {e:.3f}s were not read", "packets")
+    k1 = next((i for i in keyidx if packets[i][0] >= first - 1e-6), None)
+    if k1 is None or packets[k1][0] > last + 1e-6:
+        return [{"kind": "reencode", "start": first, "end": end_bound, "frames": len(seg)}]
+    if gop_is_open(packets, k1):
+        raise SmartFallback(f"the keyframe at {packets[k1][0]:.3f}s opens its GOP (frames decoded after it are shown "
+                            "before it), so a copy cannot start there", "open_gop")
+    k1_pts = packets[k1][0]
+    # where the copy stops: the last keyframe presented at or before the end bound, decoded after K1
+    stops = [i for i in keyidx if packets[i][1] > packets[k1][1] + 1e-9
+             and (end_bound is None or packets[i][0] <= end_bound + 1e-6)]
+    k2 = stops[-1] if stops else None
+    if k2 is not None and gop_is_open(packets, k2):
+        raise SmartFallback(f"the keyframe at {packets[k2][0]:.3f}s opens its GOP, so the copy cannot stop there",
+                            "open_gop")
+    copy_end = packets[k2][0] if k2 is not None else None
+    if copy_end is None and not to_end:
+        # no keyframe after K1 inside the segment: the copy would run past the end in its last GOP
+        return [{"kind": "reencode", "start": first, "end": end_bound, "frames": len(seg)}]
+    parts: List[Dict[str, Any]] = []
+    if first < k1_pts - 1e-6:
+        parts.append({"kind": "reencode", "start": first, "end": k1_pts,
+                      "frames": len(frames_between(frames, first, k1_pts))})
+    parts.append({"kind": "copy", "start": k1_pts, "end": copy_end, "key_dts": packets[k1][1],
+                  "end_dts": packets[k2][1] if k2 is not None else None})
+    if copy_end is not None and (end_bound is None or copy_end < end_bound - 1e-6):
+        tail = frames_between(frames, copy_end, end_bound)
+        if tail:
+            parts.append({"kind": "reencode", "start": copy_end, "end": end_bound, "frames": len(tail)})
+    return parts
+
+
+def _framehashes(cmd: List[str], code: str) -> List[Tuple[int, str]]:
+    """(size, md5) per line of an ffmpeg -f framemd5 run (packets with -c copy, frames otherwise).
+    A run that failed is no evidence, even when the hashes it printed match: a classified failure
+    (a full disk) fails the run, any other is a fallback with `code`."""
+    proc = run(cmd, quiet=True, check=False)
+    fail_if_classified(proc)
+    if proc.returncode != 0:
+        lines = [ln.strip() for ln in (proc.stderr or "").splitlines() if ln.strip()]
+        raise SmartFallback(f"the check could not hash {cmd[cmd.index('-i') + 1]}: "
+                            f"{lines[-1] if lines else f'exit code {proc.returncode}'}", code)
+    out = []
+    for line in (proc.stdout or "").splitlines():
+        if not line or line.startswith("#"):
+            continue
+        cols = [c.strip() for c in line.split(",")]
+        if len(cols) >= 6:
+            try:
+                out.append((int(cols[4]), cols[5]))
+            except ValueError:
+                continue
+    return out
+
+
+_NORMALIZE = "h264_mp4toannexb,filter_units=remove_types=7|8|9"
+
+
+def packet_hashes(path: str, seek: Optional[float] = None, t: Optional[float] = None) -> List[Tuple[int, str]]:
+    """The first video stream's packets as (size, md5) with SPS, PPS and delimiters removed."""
+    ff = require_tool("ffmpeg")
+    cmd = [ff, "-hide_banner", "-loglevel", "error", "-nostdin"]
+    if seek is not None:
+        cmd += ["-ss", f"{seek:.6f}"]
+    cmd += ["-i", path]
+    if t is not None:
+        cmd += ["-t", f"{t:.6f}"]
+    return _framehashes(cmd + ["-map", "0:v:0", "-c", "copy", "-bsf:v", _NORMALIZE, "-f", "framemd5", "-"], "packets")
+
+
+def frame_hashes(path: str, at: Optional[float], count: int) -> List[Tuple[int, str]]:
+    """framemd5 of `count` decoded frames from `at` (seconds, an accurate seek; None from the start)."""
+    ff = require_tool("ffmpeg")
+    cmd = [ff, "-hide_banner", "-loglevel", "error", "-nostdin"]
+    if at is not None and at > 0:
+        cmd += ["-ss", f"{at:.6f}"]
+    return _framehashes(cmd + ["-i", path, "-map", "0:v:0", "-frames:v", str(count), "-flags:v", "+bitexact",
+                               "-fflags", "+bitexact", "-f", "framemd5", "-"], "decode")
+
+
+def output_pts(path: str) -> List[float]:
+    proc = run([require_tool("ffprobe"), "-v", "error", "-select_streams", "v:0", "-show_entries", "packet=pts_time",
+                "-of", "csv=p=0", path], quiet=True, check=False)
+    out = []
+    for line in (proc.stdout or "").split():
+        try:
+            out.append(float(line.strip(",")))
+        except ValueError:
+            continue
+    return sorted(out)
+
+
+def _write(cmd: List[str], what: str) -> None:
+    """Run one of the smart cut's ffmpeg writes. A failure the runner classifies (a full disk, a
+    missing input) fails the run as it would anywhere; any other is a fallback with ffmpeg's reason
+    (x264 refusing the matched profile at --quality 0, a copy the muxer rejects)."""
+    proc = run(cmd, check=False)
+    fail_if_classified(proc)
+    if proc.returncode != 0:
+        lines = [ln.strip() for ln in (proc.stderr or "").splitlines() if ln.strip()]
+        why = next((ln for ln in lines if "rror" in ln), lines[-1] if lines else f"exit code {proc.returncode}")
+        raise SmartFallback(f"ffmpeg could not write {what}: {why}", "encode")
+
+
+def _count(n_frames: int, rate: Fraction) -> str:
+    """A part's exact length for the concat list's `duration`, in seconds."""
+    return f"{float(Fraction(n_frames) / rate):.9f}"
+
+
+def execute(src: str, segments: List[Tuple[float, float]], dst: str, meta: Dict[str, Any], params: Dict[str, Any],
+            packets, crf: int, preset: str, tmp: str, *, vend: Optional[float], gop_is_open, landing,
+            origin, declick: float, seek_margin: float) -> Dict[str, Any]:
+    """Write the smart cut of `segments` to `dst` and verify it (module docstring). Returns the
+    `smart` document; raises SmartFallback with the reason when it cannot be made or does not verify.
+    `landing(src, t)` is cut.seek_keyframe, `origin(path)` cut.file_origin."""
+    rate = _rate(params)
+    frame = float(1 / rate)
+    half = frame / 2
+    x264 = x264_line(params, crf, preset)
+    plans = [plan_segment(packets, s, e, vend, gop_is_open) for s, e in segments]
+    has_a = bool(meta.get("audio"))
+    for si, ((s, _), segment_parts) in enumerate(zip(segments, plans)):
+        # the sound is trimmed on the picture's grid from the segment's first frame; a picture that
+        # starts later than a frame after the segment (video that starts after its audio) would
+        # lose the sound before it, which the full re-encode keeps
+        late = segment_parts[0]["start"] - s
+        if has_a and late > frame + 1e-6:
+            raise SmartFallback(f"the picture of segment {si + 1} starts {late:.3f}s after it does, and the sound "
+                                "before it is kept only by a full re-encode", "av_offset")
+    parts: List[Dict[str, Any]] = []
+    for si, segment_parts in enumerate(plans):
+        for part in segment_parts:
+            part = dict(part, segment=si, path=os.path.join(tmp, f"part{len(parts):03d}.ts"))
+            if part["kind"] == "reencode":
+                m = min(seek_margin, part["start"])
+                trim = f"trim=start={m - half:.6f}" + (f":end={m + part['frames'] * frame - half:.6f}")
+                _write(ffmpeg_base() + ["-ss", f"{part['start'] - m:.6f}", "-i", src, "-map", "0:v:0", "-an", "-sn", "-dn",
+                                        "-vf", f"{trim},setpts=PTS-STARTPTS", "-frames:v", str(part["frames"]),
+                                        "-fps_mode", "passthrough"] + x264 + ["-f", "mpegts", part["path"]],
+                       f"the re-encoded part from {part['start']:.3f}s")
+            else:
+                target = part["start"] + _SEEK_PAST
+                landed = None if STATE.dry_run else landing(src, target)
+                if not STATE.dry_run and (landed is None or abs(landed[0] - part["start"]) > 1e-4):
+                    raise SmartFallback(f"a stream copy from the keyframe at {part['start']:.3f}s lands at "
+                                        f"{'an unreadable place' if landed is None else f'{landed[0]:.3f}s'}", "seek")
+                part["seek"] = target
+                part["t"] = None if part["end_dts"] is None else part["end_dts"] - target - _DTS_MARGIN
+                _write(ffmpeg_base() + ["-ss", f"{target:.6f}", "-i", src] + ([] if part["t"] is None else ["-t", f"{part['t']:.6f}"])
+                       + ["-map", "0:v:0", "-an", "-sn", "-dn", "-c", "copy", "-f", "mpegts", part["path"]],
+                       f"the copied part from {part['start']:.3f}s")
+            if not STATE.dry_run:
+                written = output_pts(part["path"])
+                if part["kind"] == "copy":
+                    expected = None if part["end"] is None else round((part["end"] - part["start"]) * rate)
+                    if expected is not None and len(written) != expected:
+                        raise SmartFallback(f"the copy from {part['start']:.3f}s holds {len(written)} frames where the "
+                                            f"source has {expected}", "copy_count")
+                    part["frames"] = len(written)
+                elif len(written) != part["frames"]:
+                    raise SmartFallback(f"the re-encode from {part['start']:.3f}s holds {len(written)} frames, not "
+                                        f"{part['frames']}", "encode_count")
+                if part["kind"] == "reencode":
+                    got = stream_params(part["path"]) or {}
+                    diff = mismatches(params, got)
+                    if diff:
+                        raise SmartFallback("the re-encoded part does not match the source: " + ", ".join(diff), "params_mismatch")
+            parts.append(part)
+    listfile = os.path.join(tmp, "smart.txt")
+    with open(listfile, "w", encoding="utf-8") as fh:
+        for part in parts:
+            fh.write(concat_list_line(part["path"]) + "\n")
+            if part.get("frames"):
+                fh.write(f"duration {_count(part['frames'], rate)}\n")
+    ext = os.path.splitext(dst)[1].lower()
+    cmd = ffmpeg_base() + ["-f", "concat", "-safe", "0", "-i", listfile]
+    graph: List[str] = []
+    if has_a:
+        # the sound of each segment from the source, on the picture's frame grid, encoded once
+        pads = ""
+        for i, (s, e) in enumerate(segments):
+            first = plans[i][0]["start"]
+            last = plans[i][-1]
+            end = last["end"] if last["end"] is not None else e
+            m = min(seek_margin, s)
+            cmd += ["-ss", f"{s - m:.6f}", "-t", f"{m + (end - s) + 1.0:.6f}", "-i", src]
+            a, b = m + (first - s), m + (end - s)
+            fade = min(declick, (b - a) / 2)
+            fades = (f",afade=t=in:st=0:d={fade:.6f}" if fade > 0 and i > 0 else "")
+            fades += (f",afade=t=out:st={b - a - fade:.6f}:d={fade:.6f}" if fade > 0 and i < len(segments) - 1 else "")
+            # padded to its picture's length: sound that ends early (a track shorter than the
+            # picture) must not pull the next segment's sound forward
+            graph.append(f"[{i + 1}:a:0]atrim=start={a:.6f}:end={b:.6f},asetpts=PTS-{a:.6f}/TB,"
+                         f"aresample=async=1:first_pts=0,apad=whole_dur={b - a:.6f}{fades}[a{i}]")
+            pads += f"[a{i}]"
+        graph.append(f"{pads}concat=n={len(segments)}:v=0:a=1[a]")
+        cmd += ["-filter_complex", ";".join(graph), "-map", "0:v:0", "-map", "[a]"] + aac_args()
+    else:
+        cmd += ["-map", "0:v:0"]
+    cmd += ["-c:v", "copy", "-bsf:v", "filter_units=remove_types=9", "-sn", "-dn"]
+    tb = str(params.get("time_base") or "")
+    if ext in (".mp4", ".mov", ".m4v"):
+        if tb.startswith("1/"):
+            cmd += ["-video_track_timescale", tb[2:]]
+        cmd += ["-tag:v", "avc1", "-movflags", "+faststart"]
+    _write(cmd + [dst], "the joined parts")
+    doc: Dict[str, Any] = {
+        "applied": True,
+        "parts": [{"segment": p["segment"] + 1, "kind": p["kind"], "start": round(p["start"], 6),
+                   "end": None if p["end"] is None else round(p["end"], 6), "frames": p.get("frames")} for p in parts],
+        "reencoded_frames": sum(p.get("frames") or 0 for p in parts if p["kind"] == "reencode"),
+        "copied_frames": sum(p.get("frames") or 0 for p in parts if p["kind"] == "copy"),
+        "matched": {f: params.get(f) for f in MATCH_FIELDS},
+        "encoder": x264,
+    }
+    if STATE.dry_run:
+        doc["checks"] = None
+        return doc
+    doc["checks"] = verify(src, dst, parts, params, rate, origin)
+    doc["time_base"] = doc["checks"]["time_base"]
+    return doc
+
+
+def verify(src: str, dst: str, parts: List[Dict[str, Any]], params: Dict[str, Any], rate: Fraction, origin) -> Dict[str, Any]:
+    """The checks in the module docstring; raises SmartFallback on the first that fails."""
+    frame = float(1 / rate)
+    half = frame / 2
+    pts = output_pts(dst)
+    expected = sum(p["frames"] for p in parts)
+    if len(pts) != expected:
+        raise SmartFallback(f"the output holds {len(pts)} frames where the parts hold {expected}", "frames")
+    worst = max((abs(t - pts[0] - i * frame) for i, t in enumerate(pts)), default=0.0)
+    if worst >= half:
+        at = next(i for i, t in enumerate(pts) if abs(t - pts[0] - i * frame) >= half)
+        raise SmartFallback(f"frame {at} of the output is presented at {pts[at] - pts[0]:.6f}s, off the "
+                            f"{frame:.6f}s grid (a gap or a repeated frame at a join)", "frames")
+    out_packets = packet_hashes(dst)
+    if len(out_packets) != expected:
+        raise SmartFallback(f"the output holds {len(out_packets)} video packets where the parts hold {expected}", "packets")
+    dst_origin = origin(dst)
+    offset, copied, decoded = 0, 0, 0
+    for p in parts:
+        n = p["frames"]
+        at_out = pts[offset] - dst_origin - half
+        if p["kind"] == "copy":
+            source = packet_hashes(src, p["seek"], p["t"])
+            if source != out_packets[offset:offset + n]:
+                bad = next((i for i, (x, y) in enumerate(zip(source, out_packets[offset:offset + n])) if x != y),
+                           min(len(source), n))
+                raise SmartFallback(f"copied packet {bad} of the part from {p['start']:.3f}s differs from the source's",
+                                    "packets")
+            copied += n
+            g = min(n, DECODE_CHECK_MAX_FRAMES)
+            want = frame_hashes(src, p["start"] - half, g)
+        else:
+            g = n
+            want = frame_hashes(p["path"], None, g)
+        got = frame_hashes(dst, at_out, g)
+        # a short list on both sides (a frame that did not decode) is not a match of g frames
+        if len(want) != g or want != got:
+            raise SmartFallback(f"the frames decoded from the output's {p['kind']} part at {pts[offset] - pts[0]:.3f}s "
+                                f"differ from the {'source' if p['kind'] == 'copy' else 'encoded part'}", "decode")
+        decoded += g
+        offset += n
+    out_params = stream_params(dst) or {}
+    if out_params.get("time_base") != params.get("time_base"):
+        raise SmartFallback(f"the output's time base is {out_params.get('time_base')}, the source's "
+                            f"{params.get('time_base')}", "time_base")
+    diff = mismatches(params, out_params)
+    if diff:
+        raise SmartFallback("the output's stream does not match the source: " + ", ".join(diff), "params_mismatch")
+    return {"frames": expected, "grid_worst_seconds": round(worst, 6), "copied_packets_equal": copied,
+            "frames_decoded_equal": decoded, "time_base": out_params.get("time_base")}

package/scripts/background.py

@@ -4,17 +4,22 @@
 No input file: a silent, fixed-duration, exact-size clip generated entirely
 by ffmpeg's own source filters (`color` for solid, `gradients` for a
 two-colour gradient) -- for a title card background, a placeholder behind a
-logo, or a base layer for overlay.py to composite onto.
+logo, or a base layer for overlay.py to composite onto. --transparent writes a
+fully transparent canvas in an alpha line (ProRes 4444 by default, --alpha-codec),
+the base layer graphics.py and overlay.py --keep-alpha draw on.
 
 Examples:
   python3 background.py --duration 3 --width 1920 --height 1080 --color 0x101010
   python3 background.py --duration 5 --width 1080 --height 1920 --gradient 0xff6a00:0x0057ff --angle 45
+  python3 background.py --duration 4 --width 1920 --height 1080 --transparent -o canvas.mov
 """
 import argparse
 import math
 import sys
 
-from _common import add_common, apply_common, die, draws_chroma, emit, ffmpeg_base, info, parse_time, probe, rgb_chain_to_bt709, run, validate_color, video_args, X264_PRESETS, time_arg, fmt_secs
+from _common.alpha import add_alpha_args, check_alpha_args, finish_chain, report as alpha_report, video_line
+from _common.emit import note
+from _common import STATE,  add_common, apply_common, die, draws_chroma, emit, ffmpeg_base, info, parse_time, probe, rgb_chain_to_bt709, run, validate_color, video_args, X264_PRESETS, time_arg, fmt_secs
 
 
 def main() -> int:
@@ -30,6 +35,7 @@ def main() -> int:
     ap.add_argument("--angle", type=float, default=0.0, help="gradient angle in degrees (with --gradient, default 0 = left to right)")
     ap.set_defaults(crf=18)  # --quality's default (the --crf alias was removed in 2.0)
     ap.add_argument("--preset", default="medium", choices=X264_PRESETS, help="x264 preset")
+    add_alpha_args(ap, transparent=True)
     add_common(ap)
     args = ap.parse_args()
     apply_common(args)
@@ -43,8 +49,17 @@ def main() -> int:
         die(f"--width/--height must be > 0, got width={args.width} height={args.height}")
     if args.width % 2 or args.height % 2:
         die(f"--width/--height must be even (4:2:0 chroma), got width={args.width} height={args.height}")
+    alpha_codec = check_alpha_args(args, args.output)
 
-    if args.gradient:
+    if alpha_codec:
+        # a canvas with nothing on it: alpha 0 everywhere, black colour beneath (what straight and
+        # premultiplied colour agree on), in the alpha line's own pixel format
+        if args.gradient or args.color != ap.get_default("color"):
+            die("--transparent writes a canvas with nothing on it: --color/--gradient do not apply", kind="input")
+        src_filter = (f"color=c=black@0:size={args.width}x{args.height}:rate={args.fps:g},format=rgba,"
+                      + finish_chain(alpha_codec, args.alpha_mode))
+        colours = []
+    elif args.gradient:
         try:
             c0, c1 = args.gradient.split(":")
         except ValueError:
@@ -66,7 +81,7 @@ def main() -> int:
         validate_color(args.color, "--color")
         src_filter = f"color=c={args.color}:size={args.width}x{args.height}:rate={args.fps:g}"
         colours = [args.color]
-    if draws_chroma(colours, None):
+    if colours and draws_chroma(colours, None):
         # the colours are made in RGB and converted to the BT.709 the output is tagged with: left to
         # the encoder line's own conversion (or `color`'s YUV), they were BT.601 under that tag
         # (pure red decoded as (255,24,0)); a grey is the same under every matrix
@@ -74,14 +89,26 @@ def main() -> int:
 
     output = args.output
     cmd = ffmpeg_base() + ["-f", "lavfi", "-i", src_filter, "-t", f"{target:.3f}"]
-    cmd += video_args(None, args.crf, args.preset)
+    if alpha_codec:
+        cmd += video_line(alpha_codec)
+        if STATE.hw:
+            note(f"--hw: --alpha-codec {alpha_codec} stays on the CPU (no VideoToolbox encoder writes it with its alpha)")
+    else:
+        cmd += video_args(None, args.crf, args.preset)
     cmd += ["-an", output]
     run(cmd)
 
     result = probe(output, role="output")
     v = result["video"]
     info(f"wrote {output} ({fmt_secs(result['duration'])}, {v['width']}x{v['height']}, {v['fps']:g}fps)")
-    emit(output)
+    extra: dict = {}
+    if alpha_codec:
+        # read back: the written alpha must be 0 in every frame (expected: a canvas of the same
+        # frame count, all transparent)
+        frames = result["video"].get("nb_frames") or round(target * args.fps)
+        clear = None if STATE.dry_run else {"mean": 0.0, "min": 0.0, "frames": int(frames), "per_frame": [0.0] * int(frames)}
+        alpha_report(args, alpha_codec, clear, output, extra, composite=False, provenance="specification")
+    emit(output, **extra)
     return 0
 
 

package/scripts/caption.py

@@ -75,6 +75,11 @@ __all__ = ["parse_srt", "write_srt", "transcribe", "whisper_word_timings", "die_
            "text_width_em"]
 
 ALIGN = {"bottom": 2, "top": 8, "center": 5, "bottom-left": 1, "bottom-right": 3, "top-left": 7, "top-right": 9}
+# libass reads a force_style Alignment in the legacy SSA numbering (1-3 bottom, 5-7 top, 9-11
+# middle), not the v4+ numpad one a Style line uses: only the bottom row is the same number in
+# both. Through 3.2.3 the SRT burn passed the numpad value, so --position top landed mid-left,
+# center top-left, top-left top-right (#436, measured on libass 0.17.1 and FFmpeg 9.0.2).
+FORCE_STYLE_ALIGN = {"bottom": 2, "top": 6, "center": 10, "bottom-left": 1, "bottom-right": 3, "top-left": 5, "top-right": 7}
 
 TIME_RE = re.compile(
     r"^\s*(?P<a>[\d:.,@]+)\s*(?:-->|-|–|to)\s*(?P<b>[\d:.,@]+)\s+(?P<text>.+)$"  # @ = the 1.9 @fps suffix
@@ -1750,7 +1755,7 @@ def main() -> int:
             f"Outline={args.outline:g}",
             f"Shadow={args.shadow:g}",
             f"Bold={-1 if args.bold else 0}",
-            f"Alignment={ALIGN[args.position]}",
+            f"Alignment={FORCE_STYLE_ALIGN[args.position]}",
             f"MarginV={args.margin}",
         ]
         force = ",".join(style).replace("\\", "\\\\").replace("'", "\\'")