- Add validate_path_is_safe() call in _try_delete_old_file for defense-in-depth
- Create _build_ffmpeg_base_args helper to eliminate duplicated ffmpeg argument construction
- Use os.path.normcase() for cross-platform case-insensitive path comparison
- Prevents accidental deletion of newly created file on Windows/macOS
- Extract duplicated 7-line deletion block to _try_delete_old_file helper method
- When overwriting, create new file with correct extension for selected format
- Delete old file after successful creation to avoid orphaned files
- Use video file basename for audio file naming for consistency
- Apply fix to all code paths: PIL, VHS, and standard ffmpeg
- Implemented true 'overwrite last' for video node: finds most recently modified video file and overwrites it regardless of filename
- Fixed missing sanitize_json_for_export import in video_node.py
- Added sanitize_json_for_export to image_node.py for all workflow JSON exports
- Merged upstream changes: process_batched_images, validate_path_is_safe, format_ext sanitization
- Added Pillow and numpy to requirements-bot.txt and requirements-nodes.txt for better compatibility.
- Updated comments in requirements.txt for clarity on dependency usage.
Explicitly warn users that 'overwrite_last' is negated by 'add_time'/'add_date' inputs in both image and video nodes. Also includes a specific warning for video nodes regarding Discord playback issues when disabling 'add_time'.
Updated tests to verify the presence of these warnings.
- Updated `shared/path_utils.py`: `validate_path_is_safe` now walks up the directory tree to verify the first existing ancestor is not a symlink, preventing bypasses via non-existent intermediate directories.
- Updated `tests/test_symlink_attack.py`: Added regression test `test_non_existent_directory_symlink_bypass`.
- Updated `shared/path_utils.py`: `validate_path_is_safe` now checks for symlinks in parent directories by verifying realpath vs abspath mismatch.
- Updated `nodes/video_node.py`: Added `validate_path_is_safe` check to the VHS format path recalculation block to prevent bypass.
- Updated `tests/test_symlink_attack.py`: Added regression tests for parent directory symlinks and VHS format bypass.
- Implemented `validate_path_is_safe` in `shared/path_utils.py` to reject writing to symlinks.
- Applied validation in `nodes/video_node.py` and `nodes/image_node.py` before file operations.
- Added regression test `tests/test_symlink_attack.py`.
- Updated sentinel journal with new vulnerability pattern.
Uses `process_batched_images` generator in the PIL fallback path to reduce GPU-CPU transfer overhead by processing frames in batches (default 20) instead of individually.
Also adds `tests/test_pil_batch_optimization.py` to verify the batch processing logic.
- Added safety warnings for `overwrite_last` option in Image and Video nodes.
- Clarified `resize_method` dependency in Image node tooltip.
- Added actionable instructions for `github_token` setup in Base node tooltip.
- Updated `tests/test_ux_tooltips.py` to verify new tooltip content.
Adds a critical warning to the `add_time` tooltip in `DiscordSendSaveVideo`
to inform users that disabling this option can cause single-frame playback
issues on Discord.
This moves the warning from the README into the UI where users make the
configuration choice, preventing potential bugs.
Includes regression test in `tests/test_ux_tooltips.py`.
- Updated `process_batched_images` to yield batched numpy arrays (N, H, W, C) for Tensor inputs instead of individual frames.
- Updated `DiscordSendSaveVideo.save_video` to write these batches directly to `ffmpeg` via `subprocess.stdin.write`.
- This reduces the number of system calls and Python loop iterations, improving performance significantly (from ~110 FPS to ~600 FPS in benchmarks).
- Handled `pbar` updates correctly for both batched and single-frame chunks.
- Ensured backward compatibility for list inputs (e.g. pingpong).
This test requires real PyTorch for tensor iteration which isn't available
in CI due to module mocking. The test runs correctly in local dev with torch.
- test_media.py: Use Python math.log2 instead of numpy to avoid mock issues
- test_image_node_sanitization.py: Patch tensor_to_numpy_uint8 and Image.fromarray
to properly handle mocked torch tensors
- Add numpy to requirements-nodes.txt
- Create conftest.py for test configuration
- Fix test_numpy_subprocess.py to import real numpy before any mocking
- Fix test_media.py to import real numpy early
- Fix test_image_node_sanitization.py to use MockTensor instead of torch.zeros
- Added `sanitize_token_from_text` helper to `shared/discord/webhook_client.py`.
- Updated `DiscordWebhookClient._send_with_retry` to sanitize `response.text` before returning it in error details.
- Updated `nodes/image_node.py` and `nodes/video_node.py` to sanitize `response.text` before printing error messages.
- Added regression test `tests/test_webhook_security.py`.
This prevents sensitive Discord webhook tokens from being leaked in ComfyUI console logs when the Discord API returns an error (e.g. 400 Bad Request) that echoes the request URL.
