Skip to content

World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video production studio.

fix(video_compose): honor target resolution in FFmpeg compose (vertical/9:16)

An-idd035df52b26ff
diff --git a/tests/tools/test_video_compose_vertical.py b/tests/tools/test_video_compose_vertical.py
new file mode 100644
index 00000000..7eb36d5e
--- /dev/null
+++ b/tests/tools/test_video_compose_vertical.py
@@ -0,0 +1,91 @@
+"""Vertical / arbitrary-resolution support for video_compose's FFmpeg compose.
+
+Regression test for a silent-dimension bug: the compose target resolution was
+resolved from `profile` (and the documented `metadata.compose_target` hook) but
+the per-segment scale/pad filter hardcoded 1920x1080, so vertical profiles like
+`tiktok` silently produced landscape output. These tests run the real FFmpeg
+path on a tiny lavfi fixture and assert the output dimensions.
+"""
+
+from __future__ import annotations
+
+import shutil
+import subprocess
+from pathlib import Path
+
+import pytest
+
+from tools.video.video_compose import VideoCompose
+
+pytestmark = pytest.mark.skipif(
+ shutil.which("ffmpeg") is None or shutil.which("ffprobe") is None,
+ reason="ffmpeg/ffprobe not available",
+)
+
+
+def _make_clip(path: Path, w: int = 1280, h: int = 720, d: int = 2) -> None:
+ subprocess.run(
+ ["ffmpeg", "-y", "-f", "lavfi", "-i",
+ f"color=c=teal:s={w}x{h}:d={d}:r=30",
+ "-c:v", "libx264", "-crf", "28", "-pix_fmt", "yuv420p",
+ "-g", "30", "-keyint_min", "30", str(path)],
+ capture_output=True, check=True,
+ )
+
+
+def _dims(path: Path) -> tuple[int, int]:
+ out = subprocess.check_output(
+ ["ffprobe", "-v", "error", "-select_streams", "v:0",
+ "-show_entries", "stream=width,height", "-of", "csv=p=0", str(path)]
+ ).decode().strip()
+ w, h = out.split(",")
+ return int(w), int(h)
+
+
+def _edit_decisions(src: Path, metadata: dict | None = None) -> dict:
+ ed = {
+ "version": "1.0",
+ "render_runtime": "ffmpeg",
+ "cuts": [{"id": "c1", "source": str(src), "in_seconds": 0, "out_seconds": 2}],
+ }
+ if metadata:
+ ed["metadata"] = metadata
+ return ed
+
+
+def test_compose_default_is_landscape_hd(tmp_path):
+ """No profile / no target → unchanged 1920x1080 default (backward compatible)."""
+ src = tmp_path / "in.mp4"
+ _make_clip(src)
+ out = tmp_path / "out.mp4"
+ r = VideoCompose().execute(
+ {"operation": "compose", "edit_decisions": _edit_decisions(src), "output_path": str(out)}
+ )
+ assert r.success, r.error
+ assert _dims(out) == (1920, 1080)
+
+
+def test_compose_vertical_profile(tmp_path):
+ """profile='tiktok' → 1080x1920 (the bug: previously stayed 1920x1080)."""
+ src = tmp_path / "in.mp4"
+ _make_clip(src)
+ out = tmp_path / "out.mp4"
+ r = VideoCompose().execute(
+ {"operation": "compose", "edit_decisions": _edit_decisions(src),
+ "profile": "tiktok", "output_path": str(out)}
+ )
+ assert r.success, r.error
+ assert _dims(out) == (1080, 1920)
+
+
+def test_compose_target_override_cover(tmp_path):
+ """metadata.compose_target with fit='cover' → exact requested dims, cropped to fill."""
+ src = tmp_path / "in.mp4"
+ _make_clip(src)
+ out = tmp_path / "out.mp4"
+ ed = _edit_decisions(src, metadata={"compose_target": {"width": 720, "height": 1280, "fit": "cover"}})
+ r = VideoCompose().execute(
+ {"operation": "compose", "edit_decisions": ed, "output_path": str(out)}
+ )
+ assert r.success, r.error
+ assert _dims(out) == (720, 1280)
diff --git a/tools/video/video_compose.py b/tools/video/video_compose.py
index f450183e..ec14a1dc 100644
--- a/tools/video/video_compose.py
+++ b/tools/video/video_compose.py
@@ -383,8 +383,22 @@ class VideoCompose(BaseTool):
preset = inputs.get("preset", "medium")
profile_name = inputs.get("profile")
- # Resolve target resolution from profile or default
+ # Resolve target resolution + fit mode. Priority: explicit `profile`
+ # arg > edit_decisions.metadata.compose_target > default (landscape HD).
+ # compose_target = {"width": W, "height": H, "fit": "pad"|"cover"} lets a
+ # caller request vertical (9:16) or any aspect without a named profile.
+ # fit="pad" letterboxes (no content loss, the historical default);
+ # fit="cover" scales-to-fill and centre-crops (better for vertical social).
resolution = "1920x1080"
+ fit_mode = "pad"
+ compose_target = (edit_decisions.get("metadata") or {}).get("compose_target")
+ if isinstance(compose_target, dict):
+ try:
+ resolution = f"{int(compose_target['width'])}x{int(compose_target['height'])}"
+ except (KeyError, ValueError, TypeError):
+ pass
+ if compose_target.get("fit") in ("pad", "cover"):
+ fit_mode = compose_target["fit"]
if profile_name:
try:
from lib.media_profiles import get_profile
@@ -392,6 +406,10 @@ class VideoCompose(BaseTool):
resolution = f"{p.width}x{p.height}"
except (ImportError, ValueError):
pass
+ try:
+ target_w, target_h = (int(v) for v in resolution.split("x"))
+ except ValueError:
+ target_w, target_h = 1920, 1080
cuts = edit_decisions.get("cuts", [])
if not cuts:
@@ -468,17 +486,22 @@ class VideoCompose(BaseTool):
# pix_fmt / sar across ALL segments — otherwise it throws
# "Non-monotonous DTS" or silently produces corrupt output.
#
- # Default target is 1920x1080 @ 30fps, yuv420p, sar=1. If the
- # source is smaller it letterboxes; if larger it downscales.
- # Callers can override via edit_decisions.metadata.compose_target
- # (future extension) but the defaults match the most common
- # delivery profile (YouTube landscape).
- vf_parts: list[str] = [
- "scale=1920:1080:force_original_aspect_ratio=decrease",
- "pad=1920:1080:(ow-iw)/2:(oh-ih)/2:color=black",
- "setsar=1",
- "fps=30",
- ]
+ # Target is target_w x target_h @ 30fps, yuv420p, sar=1
+ # (default 1920x1080; overridable via `profile` or
+ # edit_decisions.metadata.compose_target — see above).
+ # fit="pad" letterboxes to preserve all content; fit="cover"
+ # scales-to-fill then centre-crops (no bars, for vertical social).
+ if fit_mode == "cover":
+ geom = [
+ f"scale={target_w}:{target_h}:force_original_aspect_ratio=increase",
+ f"crop={target_w}:{target_h}",
+ ]
+ else:
+ geom = [
+ f"scale={target_w}:{target_h}:force_original_aspect_ratio=decrease",
+ f"pad={target_w}:{target_h}:(ow-iw)/2:(oh-ih)/2:color=black",
+ ]
+ vf_parts: list[str] = [*geom, "setsar=1", "fps=30"]
af_parts: list[str] = []
if speed != 1.0:
vf_parts.append(f"setpts={1.0/speed}*PTS")