siftlode/backend/app/downloads/edit.py
npeter83 f0fc542260 feat(downloads): editor backend — trim/crop derivatives (phase 2)
Per-user trim/crop clips derived from a finished download via ffmpeg, riding the same
MediaAsset/DownloadJob/worker/quota/GC machinery (asset source_kind='edit', per-user cache key):
- migration 0041: download_jobs job_kind/source_job_id/source_asset_id/edit_spec
- app/downloads/edit.py: spec normalize+sig, ffmpeg trim/crop cmd (fast stream-copy vs frame-
  accurate re-encode, per-edit 'accurate' toggle), progress+cooperative-cancel runner, filmstrip
- service.enqueue_edit; worker _process_edit branch; storage.edit_rel_path (.edits/{user} tree)
- routes: POST /edit, GET /{id}/storyboard(.jpg)
2026-07-04 01:01:46 +02:00

260 lines
9.9 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

"""Video editor (phase 2): per-user trim/crop derivatives via ffmpeg.
An edit derives a NEW per-user clip from an already-downloaded ready asset. It is never shared-
cached across users (the cache key embeds the user id), so it fully counts against the owner's
quota — but it still rides the same `MediaAsset`/`DownloadJob`/worker machinery as a download,
with the worker branching on the asset's `source_kind` to run ffmpeg instead of yt-dlp.
* trim, fast → stream copy (`-c copy`), instant, no quality loss, cuts SNAP to keyframes
(the in/out point can be off by a second or two)
* trim, accurate → re-encode (libx264 + aac), frame-accurate; cost scales with CLIP length
(the `-ss` seeks first), so a short clip is cheap even from a long source
* crop (± trim) → always re-encode (a filter can't be stream-copied), frame-accurate
`edit_spec = {trim?: {start_s, end_s}, crop?: {x, y, w, h}, accurate?: bool}`. `accurate` is the
per-edit user choice for a trim-only cut (ignored when a crop forces a re-encode anyway). `split`
is a frontend concept: the
UI fans a split out into N trim jobs, so there is one output file per job here.
Also builds the editor **filmstrip** — a single tiled sprite of the source video used as a visual
scrub track. Cached on the source asset's `storyboard_path` (reserved in phase 1) and reusable
for a player hover-preview later.
"""
import hashlib
import json
import logging
import math
import subprocess
from pathlib import Path
log = logging.getLogger("siftlode.edit")
class EditAborted(Exception):
"""Raised by run_ffmpeg when the job was paused/canceled mid-encode."""
# --- edit spec ------------------------------------------------------------------------------
def normalize_edit_spec(spec: dict | None) -> dict:
"""Canonicalize a raw edit spec: clamp/round trim, coerce crop to ints, drop empties.
The result is what gets hashed (so two equivalent specs share a cache row) and stored."""
spec = spec or {}
out: dict = {}
trim = spec.get("trim") or {}
if trim:
start = max(0.0, float(trim.get("start_s") or 0.0))
end_raw = trim.get("end_s")
t: dict = {"start_s": round(start, 3)}
if end_raw is not None:
end = float(end_raw)
if end > start:
t["end_s"] = round(end, 3)
# A trim with only a start (open-ended) is valid (cut to the end).
if t.get("start_s") or "end_s" in t:
out["trim"] = t
crop = spec.get("crop") or {}
if crop:
try:
c = {k: int(round(float(crop[k]))) for k in ("x", "y", "w", "h")}
except (KeyError, TypeError, ValueError):
c = None
if c and c["w"] > 0 and c["h"] > 0 and c["x"] >= 0 and c["y"] >= 0:
out["crop"] = c
# `accurate` only changes behaviour for a trim-only cut (a crop always re-encodes). Default to
# frame-accurate — an editor should cut where you asked; the user opts into fast/keyframe.
if out.get("trim") and "crop" not in out:
out["accurate"] = bool(spec.get("accurate", True))
return out
def edit_sig(spec: dict) -> str:
"""Stable 24-char signature of a (normalized) edit spec — the per-user cache identity."""
payload = json.dumps(normalize_edit_spec(spec), sort_keys=True, separators=(",", ":"))
return hashlib.sha256(payload.encode("utf-8")).hexdigest()[:24]
def has_crop(spec: dict) -> bool:
return bool(spec.get("crop"))
def needs_reencode(spec: dict) -> bool:
"""A crop always re-encodes; a trim re-encodes only when the user asked for an accurate cut."""
return bool(spec.get("crop")) or bool(spec.get("accurate"))
def clip_duration(spec: dict, source_duration: int | None) -> int | None:
"""Duration of the derived clip (seconds), for display + progress totals."""
trim = spec.get("trim")
if not trim:
return source_duration
start = trim.get("start_s") or 0.0
end = trim.get("end_s")
if end is None:
end = source_duration
if end is None:
return None
return max(0, round(end - start))
# --- ffmpeg: trim / crop --------------------------------------------------------------------
def build_edit_cmd(src: Path, dest: Path, spec: dict, out_ext: str) -> list[str]:
"""ffmpeg command for a trim/crop edit. `-progress pipe:1` streams machine-readable progress.
Trim uses fast input seek (`-ss` before `-i`) + `-t duration`, which is unambiguous across
ffmpeg versions. Trim-only stream-copies; a crop forces a re-encode (a filter can't be
stream-copied)."""
trim = spec.get("trim") or {}
start = trim.get("start_s") or 0.0
end = trim.get("end_s")
crop = spec.get("crop")
reencode = needs_reencode(spec)
cmd = [
"ffmpeg", "-y", "-hide_banner", "-loglevel", "error", "-nostdin",
"-progress", "pipe:1", "-nostats",
]
if start:
cmd += ["-ss", f"{start:.3f}"]
cmd += ["-i", str(src)]
if end is not None:
dur = end - start
if dur > 0:
cmd += ["-t", f"{dur:.3f}"]
if reencode:
# Input -ss + re-encode is frame-accurate (decodes from the prior keyframe, discards up to
# the exact cut). A crop adds the filter; an accurate trim re-encodes with no filter.
if crop:
cmd += ["-vf", f"crop={crop['w']}:{crop['h']}:{crop['x']}:{crop['y']}"]
cmd += [
"-c:v", "libx264", "-preset", "veryfast", "-crf", "20",
"-c:a", "aac", "-b:a", "192k",
]
else:
cmd += ["-c", "copy", "-avoid_negative_ts", "make_zero"]
if out_ext in ("mp4", "m4a", "mov"):
cmd += ["-movflags", "+faststart"]
cmd += [str(dest)]
return cmd
def _parse_out_time(line: str) -> float | None:
"""Parse a `-progress` line into elapsed output seconds."""
if line.startswith("out_time_us="):
try:
return int(line.split("=", 1)[1]) / 1_000_000
except ValueError:
return None
if line.startswith("out_time_ms="): # (some builds mislabel this as microseconds)
try:
return int(line.split("=", 1)[1]) / 1_000_000
except ValueError:
return None
return None
def run_ffmpeg(cmd, total_s, on_progress, should_cancel) -> None:
"""Run an ffmpeg edit, reporting 0100 progress and honouring cooperative cancel.
Raises EditAborted if should_cancel() turns True mid-run, RuntimeError on a nonzero exit."""
proc = subprocess.Popen(
cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True, bufsize=1
)
try:
assert proc.stdout is not None
for line in proc.stdout:
line = line.strip()
if should_cancel():
proc.terminate()
try:
proc.wait(timeout=5)
except subprocess.TimeoutExpired:
proc.kill()
raise EditAborted
secs = _parse_out_time(line)
if secs is not None and total_s:
on_progress(max(0.0, min(99.0, secs * 100.0 / total_s)))
finally:
if proc.stdout:
proc.stdout.close()
ret = proc.wait()
if ret != 0:
err = (proc.stderr.read() if proc.stderr else "") or ""
raise RuntimeError(f"ffmpeg failed ({ret}): {err[:300]}")
# --- filmstrip (editor scrub track) ---------------------------------------------------------
_SB_COLS = 12
_SB_TILE_W = 160
def storyboard_geometry(duration_s: int | None) -> dict:
"""Deterministic sprite geometry for a source duration (so meta is recomputable, unstored).
~1 frame / 5 s, clamped to fill a 12-wide grid of 24120 tiles."""
dur = max(1, int(duration_s or 1))
count = max(24, min(120, round(dur / 5)))
rows = math.ceil(count / _SB_COLS)
count = _SB_COLS * rows # fill the grid exactly
return {
"cols": _SB_COLS,
"rows": rows,
"count": count,
"interval_s": dur / count,
"tile_w": _SB_TILE_W,
}
def build_storyboard_cmd(src: Path, dest: Path, duration_s: int | None, geom: dict) -> list[str]:
"""One-pass tiled sprite. fps picks `count` frames evenly across the whole clip."""
dur = max(1, int(duration_s or 1))
fps = geom["count"] / dur
vf = f"fps={fps:.6f},scale={_SB_TILE_W}:-2,tile={geom['cols']}x{geom['rows']}"
return [
"ffmpeg", "-y", "-hide_banner", "-loglevel", "error", "-nostdin",
"-i", str(src), "-frames:v", "1", "-q:v", "5", "-vf", vf, str(dest),
]
def ensure_storyboard(asset, download_root: str, timeout_s: int = 90) -> dict | None:
"""Return the filmstrip meta for a ready source asset, generating the sprite on first use.
Best-effort + time-bounded: for a very long/high-res source the one-pass decode can be slow,
so it's capped by `timeout_s`; on timeout/failure we return None and the editor falls back to
a plain timeline (no filmstrip). Caches the sprite path on `asset.storyboard_path`.
Returns None (caller should NOT commit) if generation didn't happen; otherwise a meta dict and
sets `asset.storyboard_path` (caller commits)."""
if not asset.rel_path:
return None
geom = storyboard_geometry(asset.duration_s)
root = Path(download_root)
rel = f".storyboards/asset-{asset.id}.jpg"
dest = root / rel
if asset.storyboard_path and (root / asset.storyboard_path).exists():
return {**geom, "rel": asset.storyboard_path}
src = root / asset.rel_path
if not src.exists():
return None
dest.parent.mkdir(parents=True, exist_ok=True)
cmd = build_storyboard_cmd(src, dest, asset.duration_s, geom)
try:
subprocess.run(cmd, capture_output=True, timeout=timeout_s, check=True)
except (subprocess.TimeoutExpired, subprocess.CalledProcessError, OSError) as exc:
log.info("storyboard gen skipped for asset %s: %s", asset.id, str(exc)[:120])
return None
if not dest.exists():
return None
asset.storyboard_path = rel # caller commits
return {**geom, "rel": rel}