VanityFilter v2.8 is now available. This release improves how the app handles skin near the ears and makes exporting more reliable.
Available now
VanityFilter 2.8.0 is available now for Windows 10 version 1809 or later and Windows 11, 64-bit. Both the English and Spanish manuals have been updated to match.
Existing license formats, preset keys, and CLI interfaces are retained. The installer removes obsolete 2.7 bundle files while preserving saved preferences and activation.
Ear protection on turned faces
The ear guard now follows head tilt and keeps visible nose and cheek areas on turned faces. It is applied after feathering and temporal blending, so those stages cannot spill the mask back onto protected ear pixels. The result is that smoothing and blemish removal stop treating ears too heavily when a face rotates, while your rest-of-face settings stay as you set them.
The guard requires face landmarks. When only a five-point full-profile fallback is available, the previous coverage is retained. Blemish repair is constrained by the full-resolution mask after blur and downscaling, and pixels explicitly excluded by the actual mask are preserved through each processing pass.
Easier exports and remembered preferences
The completion dialog and the Export panel now include Play result and Open output folder actions. Export preferences persist across sessions: codec, destination folder, mask option, and sidecar option are remembered.
During preflight and export, the input, output, mask, and sidecar paths are reserved across the full batch, and both Open controls are locked. That locking prevents changing loaded inputs during processing.
Conflict checks and finalization
Temporary sibling export files are used, and all encoders complete and the sidecar is prepared before final files are replaced. Cancellation or encoding failures before finalization preserve earlier results. Individual files are replaced separately, so a filesystem error mid-finalization can leave a partial updated set rather than a transactional all-or-nothing result.
Performance and resource use
Disabled smoothing and detail passes are skipped, and merge masks are built incrementally rather than stacked. Informational CLI commands avoid loading the media stack. These changes reduce temporary mask memory and avoid unnecessary work, without running disabled corrections.
The benchmark run on 2026-09-10 used an i7-13700K with an RTX3080 on driver 591.86, measuring complete H.264 NVENC CLI exports including startup, preflight, processing, encoding, sidecar, and finalization. Two runs per version per workload, with a 12-frame warmup excluded. The blemish workload at 1920x1088 and 361 frames went from a 250.8 second mean in 2.7 to 169.6 seconds in 2.8, 32.4% less. Brightness-only at 1920x1088 and 144 frames went from 37.0 seconds to 25.6 seconds, 30.7% less. Shine at 1280x720 and 145 frames went from 50.5 seconds to 53.3 seconds, 5.5% more. Standard 1080 timing overlaps and reverses pairwise between runs, so treat those cases as comparable rather than faster. Footage, settings, and hardware affect results. The ear guard and blemish boundary changes mean outputs intentionally differ from 2.7, so no bit-identical claim applies.
Negative percent change means less elapsed time.
| Workload | 2.7 mean (range) | 2.8 mean (range) | Change |
|---|---|---|---|
| Standard 1920x1088, 144 frames | 34.6 s (30.2-38.9) | 33.2 s (31.6-34.8) | -3.9% |
| Blemish 1920x1088, 361 frames | 250.8 s (212.0-289.5) | 169.6 s (168.2-171.0) | -32.4% |
| Shine 1280x720, 145 frames | 50.5 s (34.5-66.5) | 53.3 s (39.2-67.4) | +5.5% |
| Brightness-only 1920x1088, 144 frames | 37.0 s (36.5-37.6) | 25.6 s (25.6-25.7) | -30.7% |
| Upscaled 4K 3840x2160, 48 frames | 42.9 s (42.2-43.6) | 41.7 s (41.5-41.8) | -3.0% |
These are desktop measurements, not a controlled lab. The clips had no audio, and ProRes, multiple-person, and long-clip cases were not benchmarked. The standard 1080 timing overlaps and reverses pairwise, so treat it as comparable. The 4K row is upscaled.
Variable frame rate and timing
The v2.8.0 timing preflight scans packets, not decoded frames, and does not preserve variable frame rate timestamps. It warns when it detects potential variable rate or when the scan is inconclusive, and it can flag possible audio drift. In the GUI you choose to cancel or continue; the CLI prints the warning and proceeds. Detection is not timing correction. If your source is variable frame rate, convert it to constant frame rate before processing. The packet scan has a 60 second timeout, and slow drives or long clips can delay startup before the warning appears.
Recovery and update reliability
The update checker reads and validates the update response and saved dates, and a failed manual update check is reported as a failure rather than a false up-to-date state.
Malformed presets, activation and license encoding issues, and preference type errors now recover through existing error handling instead of crashing. The update URL is a fixed HTTPS page, and the checker validates responses and saved dates. The translated banner notes are readable, and repeated checks and closing during a check are handled.
Packaging and documentation
Clean isolated package builds avoid unrelated workstation DLLs. The build uses a single OpenCV provider with an exact pinned dependency set, rejects unexpected or mismatched packages, and includes cryptography 50.0.0. The preset count and signing documentation were corrected, and the English and Spanish help files were updated.
Updating
Download VanityFilter 2.8.0 from the download page. Your saved preferences and activation carry over, and the installer removes obsolete 2.7 bundle files. The updated eight-page PDF manuals are available: read the user manual for setup and workflow details, and see the full changelog for the complete list of changes.