AI-powered video editor that turns raw footage and a creative brief into a polished ad using an ensemble of AI agents (Google Gemini + FFmpeg)
Python
from __future__ import annotations
import tempfile
import unittest
from pathlib import Path
from src.models.schemas import FootageIndex, Shot, WordTimestamp
from src.pipeline.preprocess import _words_for_shot, _words_to_text
from src.tools.captions import generate_ass_captions, has_words_in_window
class PreprocessWordAlignmentTests(unittest.TestCase):
def test_words_for_shot_preserves_timing_and_formats_transcript(self) -> None:
words = [
(0.00, 0.18, " I"),
(0.19, 0.36, "always"),
(0.37, 0.54, "taken"),
(0.55, 0.70, "care"),
(0.71, 0.80, "!"),
(1.20, 1.35, " Later"),
]
shot_words = _words_for_shot(words, 0.0, 0.85)
self.assertEqual(
[word.word for word in shot_words],
["I", "always", "taken", "care", "!"],
)
self.assertEqual(_words_to_text(shot_words), "I always taken care!")
class CaptionGenerationTests(unittest.TestCase):
def test_generate_ass_captions_writes_highlighted_dialogue_events(self) -> None:
shot = Shot(
source_file="/tmp/source.mov",
start_time=10.0,
end_time=12.5,
description="",
energy_level=0,
relevance_score=0.0,
transcript="I always taken care",
words=[
WordTimestamp(word="I", start=10.00, end=10.14),
WordTimestamp(word="always", start=10.15, end=10.42),
WordTimestamp(word="taken", start=10.44, end=10.66),
WordTimestamp(word="care", start=10.68, end=10.88),
],
)
footage_index = FootageIndex(
source_dir="/tmp",
shots=[shot],
total_duration=2.5,
created_at="2026-04-13T00:00:00Z",
)
with tempfile.TemporaryDirectory() as tmpdir:
index_path = Path(tmpdir) / "footage_index.json"
ass_path = Path(tmpdir) / "captions.ass"
index_path.write_text(
footage_index.model_dump_json(indent=2),
encoding="utf-8",
)
output = generate_ass_captions(
footage_index_path=str(index_path),
shot_id="/tmp/source.mov#10.0",
clip_start=10.0,
clip_end=11.0,
output=str(ass_path),
)
self.assertEqual(output, str(ass_path))
self.assertTrue(has_words_in_window(shot, 10.0, 11.0))
contents = ass_path.read_text(encoding="utf-8")
self.assertIn("Dialogue:", contents)
self.assertIn(r"{\1c&H4ACFFF&}", contents)
self.assertIn(r"{\1c&H4ACFFF&}I{\r} always taken care", contents)
self.assertIn(r"I always taken {\1c&H4ACFFF&}care{\r}", contents)
if __name__ == "__main__":
unittest.main()