diff --git a/README.md b/README.md
index 8b6c73d..d6bc798 100644
--- a/README.md
+++ b/README.md
@@ -585,7 +585,7 @@ VR Architecture Parameters:
   --vr_post_process_threshold VR_POST_PROCESS_THRESHOLD  Threshold for post_process feature: 0.1-0.3 (default: 0.2). Example: --vr_post_process_threshold=0.1
 
 Demucs Architecture Parameters:
-  --demucs_segment_size DEMUCS_SEGMENT_SIZE              Size of segments into which the audio is split, 1-100. Higher = slower but better quality (default: Default). Example: --demucs_segment_size=256
+  --demucs_segment_size DEMUCS_SEGMENT_SIZE              Segment length in seconds, 1-100. Smaller values use less memory. Use 'Default' to use the model's configured segment length (default: Default). Example: --demucs_segment_size=40
   --demucs_shifts DEMUCS_SHIFTS                          Number of predictions with random shifts, higher = slower but better quality (default: 2). Example: --demucs_shifts=4
   --demucs_overlap DEMUCS_OVERLAP                        Overlap between prediction windows, 0.001-0.999. Higher = slower but better quality (default: 0.25). Example: --demucs_overlap=0.25
   --demucs_segments_enabled DEMUCS_SEGMENTS_ENABLED      Enable segment-wise processing (default: True). Example: --demucs_segments_enabled=False
diff --git a/audio_separator/utils/cli.py b/audio_separator/utils/cli.py
index 927b7ad..d37a5c2 100755
--- a/audio_separator/utils/cli.py
+++ b/audio_separator/utils/cli.py
@@ -134,7 +134,7 @@ def main():
     vr_params.add_argument("--vr_enable_post_process", action="store_true", help=vr_enable_post_process_help)
     vr_params.add_argument("--vr_post_process_threshold", type=float, default=0.2, help=vr_post_process_threshold_help)
 
-    demucs_segment_size_help = "Size of segments into which the audio is split, 1-100. Higher = slower but better quality (default: %(default)s). Example: --demucs_segment_size=256"
+    demucs_segment_size_help = "Segment length in seconds, 1-100. Smaller values use less memory. Use 'Default' to use the model's configured segment length (default: %(default)s). Example: --demucs_segment_size=40"
     demucs_shifts_help = "Number of predictions with random shifts, higher = slower but better quality (default: %(default)s). Example: --demucs_shifts=4"
     demucs_overlap_help = "Overlap between prediction windows, 0.001-0.999. Higher = slower but better quality (default: %(default)s). Example: --demucs_overlap=0.25"
     demucs_segments_enabled_help = "Enable segment-wise processing (default: %(default)s). Example: --demucs_segments_enabled=False"
diff --git a/tests/unit/test_cli.py b/tests/unit/test_cli.py
index ebd419f..457c513 100644
--- a/tests/unit/test_cli.py
+++ b/tests/unit/test_cli.py
@@ -1,6 +1,7 @@
 import json
 import pytest
 import logging
+from pathlib import Path
 from audio_separator.utils.cli import main
 import subprocess
 import importlib.metadata
@@ -76,6 +77,21 @@ def test_cli_no_args(capsys):
     assert "Separate audio file into different stems." in captured.out
 
 
+def test_cli_demucs_segment_size_help(capsys):
+    with patch("sys.argv", ["cli.py", "--help"]):
+        with pytest.raises(SystemExit) as exc_info:
+            main()
+
+    assert exc_info.value.code == 0
+    help_output = capsys.readouterr().out
+    assert "Use 'Default' to use the model's configured segment length" in help_output
+    assert "Example: --demucs_segment_size=40" in help_output
+
+    readme = Path(__file__).parents[2].joinpath("README.md").read_text(encoding="utf-8")
+    assert "Use 'Default' to use the model's configured segment length" in readme
+    assert "Example: --demucs_segment_size=40" in readme
+
+
 # Test with multiple filename arguments
 def test_cli_multiple_filenames():
     test_args = ["cli.py", "test1.mp3", "test2.mp3"]