first stab at adding video

tests/component/test_04_library_prebake.py

@@ -0,0 +1,55 @@
+"""Phase 4 component test: library mode pre-bake of speaking-base clips.
+
+Verifies:
+- ``set_avatar`` under ``mode=library`` populates ``speaking_base_frames``
+  with ``library_base_clip_count`` entries.
+- Each cached entry has shape ``[T, H, W, 3]`` uint8.
+
+Run:
+    docker compose exec voice-chat python -m tests.component.test_04_library_prebake
+"""
+from __future__ import annotations
+
+import sys
+
+import numpy as np
+
+from server.video import VideoConfig, VideoEngine
+from tests.component._common import ensure_sample_avatar, get_logger
+
+log = get_logger("test_04")
+
+
+def run():
+    avatar_path = ensure_sample_avatar()
+    cfg = VideoConfig.from_dict(
+        {
+            "enabled": True,
+            "mode": "library",
+            "resolution": 480,
+            "fps": 16,
+            "library": {"base_clip_count": 2, "base_clip_seconds": 3},
+        }
+    )
+    engine = VideoEngine(cfg)
+
+    log.info("Loading models...")
+    engine.load_models()
+
+    log.info("Pre-baking 2 library clips...")
+    engine.set_avatar(avatar_path)
+
+    assert len(engine.speaking_base_frames) == 2, \
+        f"expected 2 base clips, got {len(engine.speaking_base_frames)}"
+    for i, frames in enumerate(engine.speaking_base_frames):
+        assert isinstance(frames, np.ndarray)
+        assert frames.ndim == 4 and frames.shape[-1] == 3
+        assert frames.dtype == np.uint8
+        log.info("  clip %d: shape=%s", i, frames.shape)
+
+    assert engine.get_idle_clip() is not None
+    log.info("PASS: library pre-bake complete")
+
+
+if __name__ == "__main__":
+    run()