Open-source AI podcast clipper. Generate vertical clips with face tracking and burned-in captions. CLI, MCP server, web app.
Python ai-toolsauto-captionscaptionscliffmpegmcpmcp-serveropusclip-alternativepodcast-clipspodcastsshortform-videovertical-videowhisper
feat: give studio an output shape (#152)
Nika Siradzeb5f398b91f89
diff --git a/backend/cli.py b/backend/cli.pyindex 27dd014..057cfc8 100644--- a/backend/cli.py+++ b/backend/cli.py@@ -431,6 +431,7 @@ def cmd_studio(args):cmd += ["--caption-style", args.caption_style,"--crop", args.crop,+ "--format", getattr(args, "format", None) or "vertical","--intro-seconds", str(args.intro_seconds),"--outro-seconds", str(args.outro_seconds),]@@ -4039,6 +4040,8 @@ def main():studio.add_argument("--assemblyai-api-key", help="AssemblyAI API key for --engine assemblyai. Prefer ASSEMBLYAI_API_KEY; command-line secrets can appear in process listings.")studio.add_argument("--caption-style", choices=["hormozi", "karaoke", "subtle", "branded"], default="hormozi")studio.add_argument("--crop", choices=["center", "face", "speaker", "speaker-hardcut"], default="face")+ studio.add_argument("--format", choices=["vertical", "horizontal", "square"], default="vertical",+ help="Output aspect ratio (default: vertical)")studio.add_argument("-o", "--output", help="Final output path")studio.add_argument("--intro-title", help="Intro headline (default: derived from first words)")studio.add_argument("--outro-title", default=None)diff --git a/backend/clip_studio.py b/backend/clip_studio.pyindex 7a2c9b9..66de96b 100644--- a/backend/clip_studio.py+++ b/backend/clip_studio.py@@ -119,13 +119,13 @@ def _find_paragraph(words: list, phrase: str) -> tuple[float, float]:return start, end-def _render_fragment(video, start, end, words, style, crop, title, out_dir):+def _render_fragment(video, start, end, words, style, crop, title, out_dir, fmt="vertical"):"""Render the fragment with face-crop + captions via the existing engine."""from services.clip_generator import generate_clip- print(f" [fragment] rendering {start:.1f}s–{end:.1f}s ({style}, crop={crop})", flush=True)+ print(f" [fragment] rendering {start:.1f}s–{end:.1f}s ({style}, crop={crop}, {fmt})", flush=True)res = generate_clip(video_path=video, start_second=start, end_second=end,- caption_style=style, crop_strategy=crop,+ caption_style=style, crop_strategy=crop, format=fmt,transcript_words=words, title=title, output_dir=out_dir,clean_fillers=True, allow_ass_fallback=True,progress_callback=lambda p, m: print(f" {p}% {m}", flush=True),@@ -133,13 +133,18 @@ def _render_fragment(video, start, end, words, style, crop, title, out_dir):return res["output_path"]-def _render_bookend(mode, title, handle, platforms, seconds, out_path, accent, bg):+def _render_bookend(mode, title, handle, platforms, seconds, out_path, accent, bg,+ width=1080, height=1920):cmd = [NODE, os.path.join(ROOT, "remotion", "render-bookend.mjs"),"--mode", mode, "--title", title,"--platforms", ",".join(platforms),"--seconds", str(seconds), "--output", out_path,"--accent", accent, "--bg", bg,+ # Sized to match the fragment. Left at the renderer's own defaults a+ # square or wide clip would get vertical cards and be letterboxed by+ # the concat below.+ "--width", str(width), "--height", str(height),]if handle:cmd += ["--handle", handle]@@ -150,12 +155,18 @@ def _render_bookend(mode, title, handle, platforms, seconds, out_path, accent, breturn out_path-def _concat(parts: list[str], out_path: str, fps: int = 30):+def _concat(parts: list[str], out_path: str, fps: int = 30,+ width: int = 1080, height: int = 1920):"""Concatenate parts (intro + clip + outro) in a single ffmpeg pass.- Every part is normalized to 1080x1920 @ fps with stereo 44.1k audio inside- the filtergraph, then joined with the concat filter. This gives an exact- summed duration (no crossfade-offset drift) and a single clean re-encode.+ Every part is normalized to width x height @ fps with stereo 44.1k audio+ inside the filtergraph, then joined with the concat filter. This gives an+ exact summed duration (no crossfade-offset drift) and a single clean+ re-encode.++ The size comes from the format spec rather than being fixed at 1080x1920,+ which is what services/formats.py exists to prevent: normalizing a square+ fragment to a vertical canvas would pillarbox the clip it just rendered."""n = len(parts)inputs = []@@ -165,10 +176,10 @@ def _concat(parts: list[str], out_path: str, fps: int = 30):fc = []labels = []for i in range(n):- # Normalize video: scale to fit 1080x1920, pad, set fps + SAR + format.+ # Normalize video: scale to fit the canvas, pad, set fps + SAR + format.fc.append(- f"[{i}:v]scale=1080:1920:force_original_aspect_ratio=decrease,"- f"pad=1080:1920:(ow-iw)/2:(oh-ih)/2:black,setsar=1,fps={fps},format=yuv420p[v{i}];"+ f"[{i}:v]scale={width}:{height}:force_original_aspect_ratio=decrease,"+ f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black,setsar=1,fps={fps},format=yuv420p[v{i}];")# Normalize audio to a common format so concat doesn't choke.fc.append(@@ -204,6 +215,8 @@ def main():ap.add_argument("--engine", choices=["whisper-py", "whispercpp", "assemblyai"], default=None, help="Transcription engine")ap.add_argument("--caption-style", default="hormozi", choices=["hormozi", "karaoke", "subtle", "branded"])ap.add_argument("--crop", default="face", choices=["center", "face", "speaker", "speaker-hardcut"])+ ap.add_argument("--format", default="vertical", choices=["vertical", "horizontal", "square"],+ help="Output aspect ratio (default: vertical)")ap.add_argument("--output", default=None, help="Final output path")# bookends (defaults are None so we can tell what the user explicitly set;# unset values fall back to the saved brand config, then BRAND_DEFAULTS)@@ -256,9 +269,15 @@ def main():else:raise SystemExit("Provide either --start/--end or --paragraph")+ # The canvas every part is rendered and stitched on. One lookup, so the+ # fragment, the bookends and the concat cannot disagree about the shape.+ from services.formats import get_format+ spec = get_format(args.format)+# 1. Fragmentfragment = _render_fragment(video, start, end, words, args.caption_style, args.crop, "fragment", out_dir,+ fmt=args.format,)platforms = [p.strip() for p in platforms_str.split(",") if p.strip()]@@ -274,6 +293,7 @@ def main():intro = _render_bookend("intro", intro_title, handle, platforms,args.intro_seconds, os.path.join(out_dir, "_intro.mp4"), accent, bg,+ width=spec.width, height=spec.height,)parts.append(intro)@@ -284,6 +304,7 @@ def main():outro = _render_bookend("outro", outro_title, handle, platforms,args.outro_seconds, os.path.join(out_dir, "_outro.mp4"), accent, bg,+ width=spec.width, height=spec.height,)parts.append(outro)@@ -297,7 +318,7 @@ def main():import shutilshutil.copy(parts[0], final)else:- _concat(parts, final)+ _concat(parts, final, width=spec.width, height=spec.height)dur = _probe_duration(final)print(f"\n ✓ DONE {final}")diff --git a/tests/test_studio_format.py b/tests/test_studio_format.pynew file mode 100644index 0000000..3d2a876--- /dev/null+++ b/tests/test_studio_format.py@@ -0,0 +1,149 @@+"""Tests for `podcli studio --format`.++The studio pipeline renders three things and stitches them: the fragment, the+two bookend cards, and the concat that joins them. All three had the vertical+canvas baked in, so the shape has to reach every one of them or the result is a+correctly-shaped fragment pillarboxed onto a 1080x1920 canvas.++These assert the wiring rather than the pixels: that the shape survives the+hand-off to the render script, that it defaults to what the command did before,+and that each stage is handed the canvas the format spec names.+"""++import argparse+import os+import sys+import unittest+from unittest import mock++ROOT = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))+BACKEND_ROOT = os.path.join(ROOT, "backend")+if BACKEND_ROOT not in sys.path:+ sys.path.insert(0, BACKEND_ROOT)++import cli as cli_mod+from services.formats import get_format+++def _studio_args(**overrides):+ """The namespace `podcli studio` builds, with only the shape worth varying."""+ args = argparse.Namespace(+ video="video.mp4", save_brand=False, start=0.0, end=10.0, paragraph=None,+ language=None, engine=None, assemblyai_api_key=None,+ caption_style="hormozi", crop="face", format="vertical",+ intro_seconds=2.0, outro_seconds=3.0, outro_title=None, platforms=None,+ accent=None, bg=None, intro_title=None, handle=None, output=None,+ no_intro=True, no_outro=True,+ )+ for key, value in overrides.items():+ setattr(args, key, value)+ return args+++def _run_studio(args):+ """Run cmd_studio with the render script stubbed, and return its argv."""+ with mock.patch("subprocess.run") as run:+ run.return_value = mock.Mock(returncode=0)+ with mock.patch.object(cli_mod.os.path, "exists", return_value=True):+ with self_exit():+ cli_mod.cmd_studio(args)+ return run.call_args[0][0]+++class self_exit:+ """cmd_studio ends by handing the script's exit code back up."""++ def __enter__(self):+ return self++ def __exit__(self, exc_type, exc, tb):+ return exc_type is SystemExit+++class StudioFormatTests(unittest.TestCase):+ def test_the_shape_reaches_the_render_script(self):+ cmd = _run_studio(_studio_args(format="square"))+ self.assertIn("--format", cmd)+ self.assertEqual(cmd[cmd.index("--format") + 1], "square")++ def test_defaults_to_the_shape_it_always_produced(self):+ cmd = _run_studio(_studio_args())+ self.assertEqual(cmd[cmd.index("--format") + 1], "vertical")++ def test_a_caller_that_names_no_shape_still_gets_one(self):+ """Older callers build this namespace without a format at all."""+ args = _studio_args()+ del args.format+ cmd = _run_studio(args)+ self.assertEqual(cmd[cmd.index("--format") + 1], "vertical")+++class StudioCanvasTests(unittest.TestCase):+ """Each stage must be handed the canvas, and it must be the spec's."""++ def test_bookend_is_rendered_at_the_canvas_it_is_given(self):+ import clip_studio++ with mock.patch.object(clip_studio.subprocess, "run") as run:+ run.return_value = mock.Mock(returncode=0, stderr="")+ with mock.patch.object(clip_studio.os.path, "exists", return_value=True):+ clip_studio._render_bookend(+ "intro", "Title", "@handle", ["tiktok"], 2.0, "/tmp/x.mp4",+ "#FFE000", "#0B0B0F", width=1080, height=1080,+ )++ cmd = run.call_args[0][0]+ self.assertEqual(cmd[cmd.index("--width") + 1], "1080")+ self.assertEqual(cmd[cmd.index("--height") + 1], "1080")++ def test_bookend_still_defaults_to_the_vertical_canvas(self):+ import clip_studio++ with mock.patch.object(clip_studio.subprocess, "run") as run:+ run.return_value = mock.Mock(returncode=0, stderr="")+ with mock.patch.object(clip_studio.os.path, "exists", return_value=True):+ clip_studio._render_bookend(+ "outro", "Title", None, ["tiktok"], 3.0, "/tmp/x.mp4",+ "#FFE000", "#0B0B0F",+ )++ cmd = run.call_args[0][0]+ self.assertEqual(cmd[cmd.index("--width") + 1], "1080")+ self.assertEqual(cmd[cmd.index("--height") + 1], "1920")++ def test_concat_normalizes_onto_the_canvas_it_is_given(self):+ import clip_studio++ with mock.patch.object(clip_studio.subprocess, "run") as run:+ run.return_value = mock.Mock(returncode=0, stderr="")+ with mock.patch.object(clip_studio.os.path, "exists", return_value=True):+ clip_studio._concat(["/a.mp4", "/b.mp4"], "/out.mp4",+ width=1920, height=1080)++ cmd = run.call_args[0][0]+ graph = cmd[cmd.index("-filter_complex") + 1]+ self.assertIn("scale=1920:1080", graph)+ self.assertIn("pad=1920:1080", graph)+ self.assertNotIn("1080:1920", graph)++ def test_concat_still_defaults_to_the_vertical_canvas(self):+ import clip_studio++ with mock.patch.object(clip_studio.subprocess, "run") as run:+ run.return_value = mock.Mock(returncode=0, stderr="")+ with mock.patch.object(clip_studio.os.path, "exists", return_value=True):+ clip_studio._concat(["/a.mp4", "/b.mp4"], "/out.mp4")++ cmd = run.call_args[0][0]+ graph = cmd[cmd.index("-filter_complex") + 1]+ self.assertIn("scale=1080:1920", graph)++ def test_every_shape_names_a_canvas(self):+ for shape in ("vertical", "horizontal", "square"):+ spec = get_format(shape)+ self.assertGreater(spec.width, 0)+ self.assertGreater(spec.height, 0)+++if __name__ == "__main__":+ unittest.main()