diff --git a/pyproject.toml b/pyproject.toml
index 55a5537..664c05f 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -10,7 +10,6 @@ dependencies = [
     "opencv-python",
     "scikit-learn",
     "textual>=2.0.0",
-    "textual-autocomplete>=4.0.0",
 ]
 
 [project.scripts]
diff --git a/src/faceblur/app.py b/src/faceblur/app.py
index 4215cf8..10392de 100644
--- a/src/faceblur/app.py
+++ b/src/faceblur/app.py
@@ -23,7 +23,48 @@ from textual.widgets import (
     Select,
     Static,
 )
-from textual_autocomplete import PathAutoComplete
+from textual.events import Key
+import glob
+
+
+class PathInput(Input):
+    """Custom input for bash-like path tab completion."""
+
+    def on_key(self, event: Key) -> None:
+        if event.key == "tab":
+            current = self.value
+            if not current:
+                return
+
+            event.prevent_default()
+            event.stop()
+
+            path = os.path.expanduser(current)
+            matches = glob.glob(path + "*")
+
+            if os.path.isdir(path) and not current.endswith("/"):
+                matches = glob.glob(path + "/*")
+                if not matches:
+                    self.value = current + "/"
+                    self.cursor_position = len(self.value)
+                    return
+
+            if not matches:
+                return
+
+            if len(matches) == 1:
+                match = matches[0]
+                if os.path.isdir(match):
+                    match += "/"
+            else:
+                match = os.path.commonprefix(matches)
+
+            if current.startswith("~"):
+                match = match.replace(os.path.expanduser("~"), "~", 1)
+
+            if match != current:
+                self.value = match
+                self.cursor_position = len(self.value)
 
 
 LOGO = r"""
@@ -47,7 +88,6 @@ class WelcomeScreen(Screen):
     #app-container {
         width: 60;
         height: auto;
-        border: heavy $accent;
         padding: 1 2;
     }
 
@@ -108,12 +148,10 @@ class WelcomeScreen(Screen):
 
                     with Horizontal(classes="form-row"):
                         yield Label("Video file:", classes="form-label")
-                        video_input = Input(
+                        yield PathInput(
                             placeholder="Enter path to video...",
                             id="video-input",
                         )
-                        yield video_input
-                        yield PathAutoComplete(target=video_input, path=".")
 
                     with Horizontal(classes="form-row"):
                         yield Label("Frame interval:", classes="form-label")
@@ -172,7 +210,6 @@ class ProcessingScreen(Screen):
     #processing-container {
         width: 60;
         height: auto;
-        border: heavy $accent;
         padding: 1 2;
     }
 
@@ -339,11 +376,18 @@ class FaceSelectionScreen(Screen):
     #selection-container {
         width: 65;
         height: auto;
-        max-height: 30;
-        border: heavy $accent;
         padding: 1 2;
     }
 
+    #faces-list {
+        height: auto;
+        max-height: 20;
+        overflow-y: auto;
+        margin-bottom: 1;
+        border: solid $accent-muted;
+        padding: 1;
+    }
+
     #selection-title {
         text-align: center;
         text-style: bold;
@@ -428,28 +472,31 @@ class FaceSelectionScreen(Screen):
                         "Uncheck faces you want to KEEP visible.",
                         id="selection-help",
                     )
-                    for cluster in self.real_clusters:
-                        sample_info = ""
-                        if cluster.id in self.face_samples:
-                            sample_info = (
-                                f"  [dim]{self.face_samples[cluster.id]}[/dim]"
-                            )
-                        with Horizontal(classes="face-row"):
-                            yield Checkbox(
-                                f"Person {cluster.id + 1}  "
-                                f"({len(cluster.faces)} detections)"
-                                f"{sample_info}",
-                                value=True,
-                                id=f"cluster-{cluster.id}",
-                                classes="face-checkbox",
-                            )
+
+                    with Vertical(id="faces-list"):
+                        for cluster in self.real_clusters:
+                            sample_info = ""
                             if cluster.id in self.face_samples:
-                                yield Button(
-                                    "View",
-                                    id=f"view-{cluster.id}",
-                                    variant="default",
-                                    classes="view-btn",
+                                sample_info = (
+                                    f"  [dim]{self.face_samples[cluster.id]}[/dim]"
                                 )
+                            with Horizontal(classes="face-row"):
+                                yield Checkbox(
+                                    f"Person {cluster.id + 1}  "
+                                    f"({len(cluster.faces)} detections)"
+                                    f"{sample_info}",
+                                    value=True,
+                                    id=f"cluster-{cluster.id}",
+                                    classes="face-checkbox",
+                                )
+                                if cluster.id in self.face_samples:
+                                    yield Button(
+                                        "View",
+                                        id=f"view-{cluster.id}",
+                                        variant="default",
+                                        classes="view-btn",
+                                    )
+
                     with Horizontal(id="blur-method-row"):
                         yield Label("Blur method:", id="blur-method-label")
                         yield Select(
@@ -542,7 +589,6 @@ class EncodingScreen(Screen):
     #encoding-container {
         width: 60;
         height: auto;
-        border: heavy $accent;
         padding: 1 2;
     }
 
@@ -614,14 +660,18 @@ class EncodingScreen(Screen):
 
     def _encode(self) -> None:
         """Run encoding in background thread."""
-        from .encode import encode_video
+        from .encode import encode_video, find_best_encoder
 
         def on_progress(current: int, total: int) -> None:
             pct = (current / total) * 100 if total > 0 else 0
-            self._update_ui(f"Frame {current}/{total}", pct)
+            self._update_ui(f"Frame {current}/{total} (Encoder: {encoder_name})", pct)
 
         try:
-            encode_video(
+            best_enc = find_best_encoder()
+            encoder_name = best_enc[0]
+            self._update_ui(f"Starting encoding with {encoder_name}...", 0)
+
+            encoder_used = encode_video(
                 input_path=self.video_path,
                 output_path=self.output_path,
                 clusters=self.clusters,
@@ -629,8 +679,9 @@ class EncodingScreen(Screen):
                 frame_interval=self.interval,
                 blur_method=self.blur_method,
                 progress_callback=on_progress,
+                encoder_override=best_enc,
             )
-            self._update_ui("Encoding complete!", 100)
+            self._update_ui(f"Encoding complete! (Used {encoder_used})", 100)
             self.app.call_from_thread(self._show_done)
         except Exception as e:
             self._update_ui(f"Error: {e}", 0)
@@ -657,6 +708,8 @@ class PyFaceBlurApp(App):
     CSS = """
     Screen {
         background: $surface;
+        border: heavy $accent;
+        padding: 1 2;
     }
     """
     BINDINGS = [
diff --git a/src/faceblur/blur.py b/src/faceblur/blur.py
index 58e9dbe..b3e985d 100644
--- a/src/faceblur/blur.py
+++ b/src/faceblur/blur.py
@@ -149,20 +149,10 @@ def get_bboxes_for_frame(
             bbox = interpolate_bboxes(prev_by_cluster[cid], next_by_cluster[cid], t)
             result.append((cid, bbox))
         elif cid in prev_by_cluster:
-            # Face leaving: shrink to center of previous bbox
-            x1, y1, x2, y2 = prev_by_cluster[cid]
-            cx, cy = (x1 + x2) // 2, (y1 + y2) // 2
-            center_bbox = (cx, cy, cx, cy)
-            # Interpolate from full bbox (t=0) to center point (t=1)
-            bbox = interpolate_bboxes(prev_by_cluster[cid], center_bbox, t)
-            result.append((cid, bbox))
+            # Face leaving: hold previous bbox static to ensure privacy until next keyframe
+            result.append((cid, prev_by_cluster[cid]))
         else:
-            # Face entering: grow from center of next bbox
-            x1, y1, x2, y2 = next_by_cluster[cid]
-            cx, cy = (x1 + x2) // 2, (y1 + y2) // 2
-            center_bbox = (cx, cy, cx, cy)
-            # Interpolate from center point (t=0) to full bbox (t=1)
-            bbox = interpolate_bboxes(center_bbox, next_by_cluster[cid], t)
-            result.append((cid, bbox))
+            # Face entering: hold next bbox static to ensure privacy from previous keyframe
+            result.append((cid, next_by_cluster[cid]))
 
     return result
diff --git a/src/faceblur/encode.py b/src/faceblur/encode.py
index f9d2da9..7524139 100644
--- a/src/faceblur/encode.py
+++ b/src/faceblur/encode.py
@@ -92,35 +92,129 @@ def probe_video(video_path: Path) -> dict:
     return info
 
 
-def find_best_encoder() -> str:
+def find_best_encoder() -> Tuple[str, List[str], List[str]]:
     """Find the best available H.264 encoder by testing each in priority order.
 
     Returns:
-        Name of the best available encoder
+        (encoder_name, input_args, output_args)
     """
-    for encoder in ENCODER_PRIORITY:
-        cmd = [
-            "ffmpeg",
-            "-v",
-            "quiet",
-            "-f",
-            "lavfi",
-            "-i",
-            "nullsrc=s=64x64:d=0.1",
-            "-c:v",
-            encoder,
-            "-f",
-            "null",
-            "-",
-        ]
-        result = subprocess.run(cmd, capture_output=True, timeout=10)
-        if result.returncode == 0:
-            return encoder
+    # 1. NVIDIA NVENC
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-c:v",
+        "h264_nvenc",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return "h264_nvenc", [], []
 
-    raise RuntimeError(
-        "No H.264 encoder found. Available encoders checked: "
-        + ", ".join(ENCODER_PRIORITY)
-    )
+    # 2. Linux VA-API (AMD/Intel)
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-vaapi_device",
+        "/dev/dri/renderD128",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-vf",
+        "format=nv12,hwupload",
+        "-c:v",
+        "h264_vaapi",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return (
+            "h264_vaapi",
+            ["-vaapi_device", "/dev/dri/renderD128"],
+            ["-vf", "format=nv12,hwupload"],
+        )
+
+    # 3. AMD AMF (Windows/Proprietary Linux)
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-c:v",
+        "h264_amf",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return "h264_amf", [], []
+
+    # 4. Intel QSV
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-c:v",
+        "h264_qsv",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return "h264_qsv", [], []
+
+    # 5. Software fallback (libx264 if installed)
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-c:v",
+        "libx264",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return "libx264", [], []
+
+    # 6. Software fallback (libopenh264)
+    cmd = [
+        "ffmpeg",
+        "-v",
+        "quiet",
+        "-f",
+        "lavfi",
+        "-i",
+        "nullsrc=s=64x64:d=0.1",
+        "-c:v",
+        "libopenh264",
+        "-f",
+        "null",
+        "-",
+    ]
+    if subprocess.run(cmd, capture_output=True, timeout=5).returncode == 0:
+        return "libopenh264", [], []
+
+    raise RuntimeError("No suitable H.264 encoder found on this system.")
 
 
 def build_keyframe_bboxes(
@@ -162,7 +256,8 @@ def encode_video(
     frame_interval: int,
     blur_method: BlurMethod = "gaussian",
     progress_callback: Optional[Callable[[int, int], None]] = None,
-) -> None:
+    encoder_override: Optional[Tuple[str, List[str], List[str]]] = None,
+) -> str:
     """Re-encode video with face blur applied to selected clusters.
 
     Args:
@@ -173,9 +268,17 @@ def encode_video(
         frame_interval: Frame interval used during detection
         blur_method: Blur method to use
         progress_callback: Called with (current_frame, total_frames)
+        encoder_override: Optional tuple of (encoder_name, enc_in_args, enc_out_args)
+
+    Returns:
+        String description of the encoder used (for UI reporting)
     """
     video_info = probe_video(input_path)
-    encoder = find_best_encoder()
+
+    if encoder_override:
+        encoder_name, enc_in_args, enc_out_args = encoder_override
+    else:
+        encoder_name, enc_in_args, enc_out_args = find_best_encoder()
 
     keyframe_bboxes, keyframe_indices = build_keyframe_bboxes(
         clusters,
@@ -203,37 +306,38 @@ def encode_video(
     height, width = first_frame.shape[:2]
 
     # Build FFmpeg encode command
-    ffmpeg_cmd = [
-        "ffmpeg",
-        "-y",
-        "-f",
-        "rawvideo",
-        "-pix_fmt",
-        "bgr24",
-        "-s",
-        f"{width}x{height}",
-        "-r",
-        str(fps),
-        "-i",
-        "pipe:0",
-        "-i",
-        str(input_path),
-        "-map",
-        "0:v:0",
-    ]
+    ffmpeg_cmd = ["ffmpeg", "-y"]
+    ffmpeg_cmd.extend(enc_in_args)
+    ffmpeg_cmd.extend(
+        [
+            "-f",
+            "rawvideo",
+            "-pix_fmt",
+            "bgr24",
+            "-s",
+            f"{width}x{height}",
+            "-r",
+            str(fps),
+            "-i",
+            "pipe:0",
+            "-i",
+            str(input_path),
+            "-map",
+            "0:v:0",
+        ]
+    )
 
     # Map audio from original if present
     if video_info["audio_codec"]:
         ffmpeg_cmd.extend(["-map", "1:a:0", "-c:a", "copy"])
 
+    ffmpeg_cmd.extend(enc_out_args)
     ffmpeg_cmd.extend(
         [
             "-c:v",
-            encoder,
+            encoder_name,
             "-b:v",
             str(bitrate),
-            "-pix_fmt",
-            "yuv420p",
             str(output_path),
         ]
     )
@@ -283,3 +387,5 @@ def encode_video(
     if proc.returncode != 0:
         stderr = proc.stderr.read().decode() if proc.stderr else ""
         raise RuntimeError(f"FFmpeg encoding failed: {stderr}")
+
+    return encoder_name
diff --git a/uv.lock b/uv.lock
index 368d3d2..e95b025 100644
--- a/uv.lock
+++ b/uv.lock
@@ -102,7 +102,6 @@ dependencies = [
     { name = "opencv-python" },
     { name = "scikit-learn" },
     { name = "textual" },
-    { name = "textual-autocomplete" },
     { name = "uniface" },
 ]
 
@@ -112,7 +111,6 @@ requires-dist = [
     { name = "opencv-python" },
     { name = "scikit-learn" },
     { name = "textual", specifier = ">=2.0.0" },
-    { name = "textual-autocomplete", specifier = ">=4.0.0" },
     { name = "uniface" },
 ]
 
@@ -743,19 +741,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/d3/be/e191c2a15da20530fde03564564e3e4b4220eb9d687d4014957e5c6a5e85/textual-8.0.0-py3-none-any.whl", hash = "sha256:8908f4ebe93a6b4f77ca7262197784a52162bc88b05f4ecf50ac93a92d49bb8f", size = 718904, upload-time = "2026-02-16T17:12:11.962Z" },
 ]
 
-[[package]]
-name = "textual-autocomplete"
-version = "4.0.6"
-source = { registry = "https://pypi.org/simple" }
-dependencies = [
-    { name = "textual" },
-    { name = "typing-extensions" },
-]
-sdist = { url = "https://files.pythonhosted.org/packages/1e/3a/80411bc7b94969eb116ad1b18db90f8dce8a1de441278c4a81fee55a27ca/textual_autocomplete-4.0.6.tar.gz", hash = "sha256:2ba2f0d767be4480ecacb3e4b130cf07340e033c3500fc424fed9125d27a4586", size = 97967, upload-time = "2025-09-24T21:19:20.213Z" }
-wheels = [
-    { url = "https://files.pythonhosted.org/packages/9f/66/ebe744d79c87f25a42d2654dddbd09462edd595f2ded715245a51a546461/textual_autocomplete-4.0.6-py3-none-any.whl", hash = "sha256:bff69c19386e2cbb4a007503b058dc37671d480a4fa2ddb3959c15ceb4aff9b5", size = 16499, upload-time = "2025-09-24T21:19:18.489Z" },
-]
-
 [[package]]
 name = "threadpoolctl"
 version = "3.6.0"