A desktop app for archiving, converting and restoring video — tape captures, camcorder footage, film scans, DVDs and anime — with the tools the restoration community already trusts, and no scripting.
VapourBox runs VapourSynth, QTGMC and FFmpeg for you: no scripts to write, no plugins to install, no terminal. Everything it needs is downloaded on first launch. It's free, open source, runs on macOS, Windows and Linux, and all processing stays on your machine.
- Converts and archives. Reads camcorder DV and AVCHD, MPEG-2 from discs and recorders, MXF, and ordinary MP4/MKV/AVI/MOV files, including formats consumer converters mishandle (NTSC DV's 4:1:1 chroma, 10–16-bit film scans). Writes H.264, H.265, ProRes or lossless FFV1, with hardware encoding where available. DVDs and
VIDEO_TSfolders are read directly, and aspect ratio is carried through correctly. - Fixes how the picture is stored. Detects interlaced and telecined sources, then deinterlaces with QTGMC or recovers the original film frames with inverse telecine (or by stripping soft-telecine flags).
- Cleans up old footage. Optional filters for tape noise, smeared colour, dust and scratches, flicker, ghosting, compression blocking, banding, camera shake and dark, flat pictures. They're off until you turn them on, and the presets turn on what a given kind of source needs.
- Colours black-and-white footage. Colourisation predicts colour for film scans, old broadcasts and home movies with AI (DDColor or DeOldify Video, each an optional download), steadied across frames, and keeps the brightness of every pixel exactly as it was, so the result can be graded in Color Correction.
- Restores faces. Face Restoration finds the faces in a soft VHS or low-resolution picture and redraws them with AI (RestoreFormer++, or GFPGAN or CodeFormer for non-commercial use) at the output resolution, steadied across frames to keep flicker down, leaving every other pixel untouched.
- Restores anime. Descale undoes the upscale most HD anime went through in production, BM3D cleans flat colour without smearing lines, and AI upscaling redraws line art at 2× or 4×: Real-ESRGAN (sharpest, fastest), Real-CUGAN (gentler, with denoising) or AnimeJaNai (trained on HD anime), each an optional download.
- Upscales live action with AI models trained on real-world footage, each an optional download: Real-ESRGAN General (fast, with a noise-removing variant), Live Action SPAN (repairs compression, blur and halos on DVD and broadcast sources while keeping colour), or Real-ESRGAN x4plus (most detailed, slow). An AI Strength control blends any AI upscale back towards a plain one.
- Converts frame rates — the broadcast standards or any rate you type in (24 → 48 or 60, say) — with motion interpolation or the RIFE AI interpolator (an optional download).
- Restores the sound too. Replaces a dead channel, removes hum, rumble, clicks, clipping and hiss, cleans up speech with the DeepFilterNet AI model, puts sound that is out of step or drifting back in sync, and evens out loudness.
- Writes subtitles from speech with Whisper AI, as an
.srtfile, an embedded track, or burnt into the picture.
- Add a source. Drop in a file, a folder, or a
VIDEO_TSfolder, or open a DVD. VapourBox reports what it found (interlaced, telecined or progressive), so you don't have to guess. - Choose the output and any processing. Pick a format, then start from a preset named after your source (VHS Cleanup, DV Camcorder Tape, Anime DVD, …) or switch filters on yourself. Filters that suit the source are marked Suggested — Noise Reduction and Dehalo too, when a quick measurement of the picture (run in the background as the video loads) finds it clearly noisy or ringing around its edges.
- Check the preview, then run. A before/after preview updates as you change settings. Queue as many files as you like and leave them to run.
| Platform | Minimum OS | File |
|---|---|---|
| macOS (Apple Silicon) | macOS 15 Sequoia | VapourBox-x.x.x-macos-arm64.dmg |
| macOS (Intel) | macOS 12 Monterey | VapourBox-x.x.x-macos-x64.dmg |
| Windows (x64) | Windows 10/11 | VapourBox-x.x.x-windows-x64-setup.exe |
| Linux (x64) | glibc 2.39 (Ubuntu 24.04, Debian 13) | VapourBox-x.x.x-x86_64.AppImage |
| Linux (arm64) | glibc 2.39 (Ubuntu 24.04, Debian 13) | VapourBox-x.x.x-aarch64.AppImage |
There is no account, no telemetry, and nothing is uploaded.
Installing on macOS
- Download the
.dmgfor your Mac (Apple menu → About This Mac):arm64for Apple Silicon,x64for Intel. - Open it and drag VapourBox to Applications.
- The first launch downloads its processing components (~130 MB on Apple Silicon, ~105 MB on Intel).
VapourBox is signed with an Apple Developer ID and notarized, so it opens without Gatekeeper warnings.
Installing on Windows
- Download the
-setup.exeinstaller and run it. - The installer isn't code-signed, so SmartScreen may say it "protected your PC". Choose More info → Run anyway.
- Launch VapourBox from the Start menu. The first launch downloads its processing components (~205 MB).
It installs for your user only (%LOCALAPPDATA%\Programs\VapourBox), so it never asks for administrator rights. A newer installer upgrades in place; uninstalling from Settings → Apps also removes the downloaded components, and leaves your presets alone.
Portable option: each release also has a .zip of the same build. Extract it anywhere you can write to (e.g. C:\VapourBox, not Program Files) and run vapourbox.exe.
Installing on Linux
- Download the
.AppImagefor your machine (x86_64for Intel/AMD,aarch64for ARM). - Make it executable:
chmod +x VapourBox-x.x.x-x86_64.AppImage(or tick Allow executing file as program in its Properties). - Run it. The AppImage carries its processing components, so it works offline from the first launch.
To get a menu entry and in-place updates, open it with Gear Lever, AppImageLauncher or AppImageUpdate.
Where an AppImage can't be mounted, use the .tar.gz of the same build: extract it and run ./vapourbox. It downloads its processing components (~265 MB on x64, ~220 MB on ARM) on first launch.
GPU features. RIFE frame interpolation and AI upscaling run on the graphics card through Vulkan: on any Apple Silicon Mac, and on Windows and Linux x64 PCs with a current graphics driver. They aren't available on Intel Macs or Linux on ARM yet. GPU deinterlacing and KNLMeans denoising use OpenCL instead; without it, deinterlacing falls back to the CPU, and NLMeans gives KNLMeans' result on the CPU. Whenever you choose a GPU feature your machine can't run, VapourBox says what's missing and how to fix it. Face Restoration and Colourisation run through the AI runtime instead: on the graphics card and Neural Engine on a Mac (Core ML), on any DirectX 12 graphics card on Windows (DirectML), and on the CPU on Linux, which is several times slower. On Windows, RIFE and the AI runtime also need the Microsoft Visual C++ Redistributable. On Intel Macs, Face Restoration, Colourisation and DeepFilterNet need macOS 13 or later.
Add-ons. A few features use third-party downloads that aren't part of VapourBox: the Whisper engine and language models for subtitles, the models for AI upscaling (anime and live action) and for RIFE frame interpolation, DeepFilterNet for speech enhancement in Audio Restoration, the face restoration and colourisation models, and the AI runtime (Microsoft's ONNX Runtime) that AI model add-ons run on. VapourBox offers to download one the first time you need it, stating its source, size and licence, and Settings → Add-ons installs or removes them at any time. AnimeJaNai is free for non-commercial use only (CC BY-NC-SA 4.0); for paid work, use Real-ESRGAN or Real-CUGAN instead. Of the face restoration models, RestoreFormer++ (the default) is free for any use; GFPGAN and CodeFormer are for non-commercial use only.
Older processors. Intel and AMD CPUs from before about 2013 (no AVX2) are supported on every platform. VapourBox checks the processor on first launch and installs a matching set of components, which runs some filters more slowly but never differently.
| Video | H.264, H.265, ProRes (Proxy to 4444 XQ), FFV1, HuffYUV or FFVHuff (lossless); hardware encoding via VideoToolbox, NVENC, Quick Sync or AMF |
| Audio | Passthrough; AAC, MP3, AC-3, Opus, FLAC or PCM; or none |
| Containers | MKV, MP4, MOV, AVI |
| Reads | .dv · .mts .m2ts · .vob .m2v .mpg .mpeg · .mxf · .avi .mov .mp4 .mkv .ts .wmv .webm .flv, DVD discs and VIDEO_TS folders |
| Source formats | 4:1:1 (NTSC DV), 4:1:0, 4:2:0/4:2:2/4:4:4 up to 16-bit, RGB, grayscale |
| Colour format | Match the source, or convert to 4:2:0 8-bit (plays everywhere), 4:2:0 10-bit (for GPU encoders), 4:2:2 8- or 10-bit, or 4:4:4 10-bit. Matching a 10-bit source makes a 10-bit file, which some players and browsers won't open |
| HDR | HDR10, HLG and Dolby Vision (profiles 7, 8 and 9) are processed and saved as HDR, and shown tone-mapped in the preview. H.265 keeps HDR10's mastering-display and light-level metadata. Dolby Vision's own dynamic metadata isn't kept, so the output plays as standard HDR. Dolby Vision Profile 5, used by some streaming services, can't be decoded and shows pink and green |
| Aspect ratio | Non-square pixels preserved, even through a resize; or convert to square pixels, force a display aspect, or letterbox to a target size |
Most sources need only a few of these, and the presets pick them for you. Filters run in this fixed order, the same order the app lists them in, whichever are switched on.
| # | Filter | What it fixes |
|---|---|---|
| 1 | Deinterlace | Combing on motion: QTGMC for interlaced video, IVTC for telecined film, Bwdif for speed |
| 2 | Edge Repair | Dirty rows and columns at the edge of a tape capture, rebuilt instead of cropped |
| 3 | Ghost Removal | A faint shifted copy of the picture from an aerial or long cable |
| 4 | Deflicker | Brightness pulsing between frames, typical of scanned cine film |
| 5 | DeScratch | Vertical scratches on scanned film |
| 6 | SpotLess | Dust, dirt and single-frame specks |
| 7 | Descale | Soft, haloed line art in anime the studio upscaled; Detect finds the original resolution |
| 8 | Noise Reduction | Grain and video noise. Motion-compensated by default, with BM3D, KNLMeans and more under advanced options |
| 9 | Chroma Denoise | Blotchy, smeared colour, common on VHS |
| 10 | Dehalo | Bright outlines and ringing around edges |
| 11 | Deblock | Blocking from heavy compression |
| 12 | Deband | Visible steps in gradients and skies |
| 13 | Anti-Aliasing | Stair-stepping on diagonal edges: daa, santiag, or SangNom (sharpest) |
| 14 | Chroma Fixes | Colour shifted off the picture, bleeding, rainbowing, dot crawl and leftover combing |
| 15 | Colourisation | Black-and-white footage, coloured by AI (DDColor or DeOldify Video, optional downloads); the brightness of every pixel is kept exactly |
| 16 | Color Correction | Brightness, contrast, saturation, levels, white balance, shadow detail and local contrast, set by hand or measured |
| 17 | Stabilize | Shake and weave from telecine, film scans or a handheld camera, and the frame-height "breathing" of pull-through film scanners (8mm, Super 8): measured from the footage, or read per frame from a factor file (reel.pitch.txt next to reel.mp4 is picked up automatically; file format in docs/PITCH_FACTOR_FILES.md) |
| 18 | Rotate / Flip | Sideways, upside-down or mirrored footage |
| 19 | Crop & Resize | Black borders, tape blanking and head-switching noise (measured for you by Detect borders), scaling and upscaling (edge-directed or AI) with SD colour converted to HD's BT.709, and borders to reach an exact frame size |
| 20 | Face Restoration | Blurry, low-resolution faces, redrawn by AI at the output resolution (optional download) |
| 21 | Sharpen | Soft pictures, sharpened after the resize so the result isn't softened again |
| 22 | Film Grain | Grain added back after denoising, so the picture doesn't look plastic |
| 23 | Frame Rate | PAL ↔ NTSC and other rate changes, or any custom rate, by motion interpolation, RIFE, or frame repetition |
| 24 | Audio Restoration | A dead channel, hum, rumble, clicks, clipping, hiss, sound out of sync or drifting, uneven loudness; speech enhancement with DeepFilterNet |
| 25 | Subtitles | Speech-to-text with Whisper AI, from the restored sound |
The order is designed so each step works on the right picture: damage and noise are removed first, the picture is framed and resized next, and only then sharpened and given grain. The frame rate is converted last, so every other filter works on real frames. Borders are added after everything else, so nothing touches them.
Every filter explains in place what it does and when to use it. Where two filters work against each other — sharpening that brings back halos Dehalo removed, for instance — the app tells you.
DVDs and folders
A disc: click the disc icon in the toolbar, or Open DVD. Pick a title from the list (duration, resolution, chapters and audio tracks are shown) and Add to Queue; it's extracted, analysed and queued like any other video.
A ripped disc: drag in a folder containing VIDEO_TS, the VIDEO_TS folder itself, or a flat rip (the VIDEO_TS.IFO and .VOB files directly in a folder). The title picker opens automatically.
A folder of videos: drag it in, or use Open Folder. It's searched recursively and everything found is queued.
Encrypted discs. Most commercial DVDs use CSS encryption. VapourBox doesn't include libdvdcss to decrypt them, but uses it if you install it:
- macOS:
brew install libdvdcss - Windows: put
libdvdcss-2.dllfrom VideoLAN next tovapourbox.exe - Linux:
sudo apt install libdvdcss2·sudo dnf install libdvdcss(RPM Fusion) ·sudo pacman -S libdvdcss
CSS decryption may be legally restricted where you live. VapourBox doesn't endorse or support it; libdvdcss is a third-party library that libdvdread loads if it's present.
Preview, timeline and trimming
- Click a thumbnail to jump there, or drag to scrub. The transport buttons step a frame at a time or jump to either end.
- Zoom the preview with the mouse wheel or pinch; double-click resets it. 1:1 shows actual pixels, the way to judge fine detail, and Fit shows the whole frame.
- The mouse wheel or pinch also zooms the timeline around the cursor, and Shift+wheel or a two-finger swipe pans it.
- Set In / Set Out process only part of the source, handy for trying settings on a short section.
- When one source frame becomes several output frames — a frame rate conversion, or double-rate deinterlacing — arrows under the preview step through them, marking the ones the conversion generated. They are exactly the frames the final video will contain.
Keyboard shortcuts
Every button with a shortcut shows it in its tooltip, and ? lists them all. Shortcuts without ⌘/Ctrl are ignored while you're typing in a text box.
| Action | macOS | Windows / Linux |
|---|---|---|
| Previous / next frame | ← / → | ← / → |
| Back / forward 10 frames | ⇧← / ⇧→ | Shift+← / Shift+→ |
| Go to start / end | Home / End | Home / End |
| Set in / out point | I / O | I / O |
| Go to in / out point | ⇧I / ⇧O | Shift+I / Shift+O |
| Clear in and out points | ⌥X | Alt+X |
| Preview at actual pixels (1:1) / fit | ⌘1 / ⌘0 | Ctrl+1 / Ctrl+0 |
| Zoom timeline in / out / to fit | = / - / \ | = / - / \ |
| Add videos | ⌘O | Ctrl+O |
| Start processing | ⌘↩ | Ctrl+Enter |
| Settings | ⌘, | Ctrl+, |
Presets
For Your Source: VHS Cleanup, SD Tape → HD (YouTube / TV), DV Camcorder Tape, PAL DVD / Broadcast, DVD IVTC, Anime DVD, 8mm / Super 8 Film Scan and Low-Light Phone / Camera Video. Pick the one that matches what you have.
Quality Only: Fast, Balanced and High Quality just deinterlace, with increasing effort, for a picture that's already clean.
Your own presets are saved alongside these, and can be exported and imported to share a setup that works for a particular camcorder or capture card. Because a preset can contain custom VapourSynth code and FFmpeg arguments, VapourBox shows you any it contains before importing, and lets you import the preset without them.
Advanced options
By default each filter shows a short, curated set of choices. Settings → General → Show advanced options (or the switch beside any filter's settings) reveals every method and parameter, including:
- all 70+ QTGMC parameters, and opt-in higher-precision deinterlacing (4:2:2 chroma, 16-bit processing);
- the less common denoisers: DFTTest, FFT3DFilter, TTempSmooth, FluxSmooth, STPresso, TemporalDegrain2, BM3D, KNLMeans (GPU), NLMeans (the same filter on the CPU) and a large-window median;
- specialist repairs: the vine dehalo, mpeg2stinx for the comb-like noise of film DVDs, and a fade-fix mode for IVTC (fades added after telecine);
- fine control of edge-directed upscaling (NNEDI3's neurons, quality and prescreener; EEDI3's weights);
- Descale's explicit native width and edge handling.
Turning advanced options off never changes what a filter does: a method a preset chose stays selected and visible.
Other settings
All under Settings → General:
- Appearance: Auto (follows the system), Light, Dark, Monokai or Pride.
- Temporary Files: where scratch files go. DVD extraction can need several GB, so move this off a small or slow drive if needed.
- Existing Output Files: ask (the default), always overwrite, or always write a new numbered file (
name (2).ext).
Shutting down after an overnight queue
While a queue runs, tick Shut down computer when the queue finishes under the progress bar. When the last video is done, even if some failed, VapourBox counts down for 60 seconds (with Cancel Shutdown and Shut Down Now), then asks the system to shut down. The option only applies to the run it was ticked for.
A failed video's log is saved first, as <output name>.error.log next to where its output would have gone.
The shutdown is never forced, so an app with unsaved work can stop it. VapourBox checks when you tick the option that the shutdown will be allowed:
- macOS asks once for permission to control System Events. If you refuse, allow it later in System Settings → Privacy & Security → Automation.
- Linux needs systemd-logind and an account allowed to power off without a password (normally true for whoever is logged in at the desktop).
docs/FEATURE_MAP.md compares VapourBox, feature by feature, with HandBrake, FastFlix, StaxRip, Hybrid, Shutter Encoder, Topaz Video, Video2X and other AI enhancers, the community VHS restoration tools, and the archival tools (vrecord, DVRescue, QCTools) that work alongside it, including where VapourBox is behind.
Settings → General → "Report a bug or give feedback" opens the issue tracker. Reports about specific problem sources are especially useful; several fixes came from them.
See docs/BUILDING.md for build instructions, project structure and the development workflow, and docs/ADDONS.md for how optional downloads (AI models and the AI runtime) are described, installed and tested.
The investigations behind feature decisions (measurements, screenshots and the reasoning, including ideas that weren't built) are written up in docs/reports/.
Free software under the GNU General Public License, version 3 or (at your option) any later version. Full licence texts and third-party attributions are in licenses/.
Stuart Cameron — stuart-cameron.com
VapourBox is a front end to other people's work. Thanks in particular to:
- QTGMC by Vit, originally based on TempGaussMC_beta2 by Didée — the deinterlacing algorithm
- VIVTC by Fredrik Mellbin — field matching and decimation for inverse telecine
- VapourSynth by Fredrik Mellbin — the video processing framework
- havsfunc, maintained by HolyWu — the script library QTGMC and many other filters come from
- MVTools — the motion estimation QTGMC is built on; VapourSynth port by dubhater, from the AviSynth plugin by Manao with later work by Fizick, Pinterf and the SVP team
- NNEDI3 by Kevin Stone (tritical) — the edge-directed interpolator behind QTGMC, via znedi3 by sekrit-twc and dubhater's nnedi3, whose NEON kernels make it fast on Apple Silicon and ARM Linux
- akarin by Akarin, now maintained by the Jaded Encoding Thaumaturgy project — the expression compiler that makes Apple Silicon and ARM Linux several times faster
- fmtconv by Firesledge (Laurent de Soras) — format conversion and resampling
- zsmooth by Austin Dworaczyk Wiltshire — chroma denoising (CCD, originally by Sergey Stolyarevsky for VirtualDub) and the RemoveGrain family
- dubhater — VapourSynth ports of MVTools, nnedi3, AWarpSharp2, TemporalMedian, FluxSmooth, Bifrost and
adjust - HolyWu — VapourSynth ports of DFTTest, TTempSmooth, TCanny, CTMF, DCTFilter, Deblock, AddGrain, CAS, EEDI3 and NNEDI3CL
- mawen1250 — Retinex, BM3D and mvsfunc
- Edoardo Brunetti (maintained by Jaded Encoding Thaumaturgy) — KNLMeansCL
- Frechdachs (maintained by Jaded Encoding Thaumaturgy) — descale
- vszip by Julek — CLAHE, the local-contrast lift in Color Correction, and NLMeans, the CPU non-local-means denoiser
- SangNom by james1201 and Victor Efimov — the SangNom anti-aliasing method
- vs-jetpack by the Jaded Encoding Thaumaturgy project — FixInterlacedFades, vine_dehalo and mpeg2stinx, ported into VapourBox's filter scripts
- RIFE by hzwer (with Megvii) — the frame interpolation network, an optional download, run through HolyWu's VapourSynth plugin (built on nihui's rife-ncnn-vulkan and Tencent's ncnn), with MoltenVK for Vulkan on macOS
- Real-ESRGAN by Xintao Wang and Real-CUGAN by bilibili — the anime upscaling networks, an optional download, run through AmusementClub's vs-mlrt (VSNCNN)
- AnimeJaNai by the-database — anime upscaling models, an optional download (CC BY-NC-SA 4.0, non-commercial use only)
- RestoreFormer++ by Zhouxia Wang, Jiawei Zhang, Tianshui Chen, Wenping Wang and Ping Luo — the default face restoration model, with the RetinaFace detector by biubug6 (Pytorch_Retinaface), the BiSeNet face parser by zll (face-parsing.PyTorch) and facexlib by Xintao Wang, an optional download
- GFPGAN by THL A29 Limited, a Tencent company, and CodeFormer by S-Lab — alternative face restoration models, optional downloads for non-commercial use only
- DDColor by Xiaoyang Kang, Tao Yang, Wenqi Ouyang, Peiran Ren, Lingzhi Li and Xuansong Xie, and DeOldify by Jason Antic — the colourisation models, optional downloads
- Real-ESRGAN general and x4plus/x2plus models by Xintao Wang, and 2xLiveActionV1_SPAN by jcj83429 — upscaling models for real-world footage, optional downloads (BSD-3-Clause; Apache-2.0)
- whisper.cpp by the ggml authors and OpenAI's Whisper models — speech recognition for subtitles, an optional download
- ONNX Runtime by Microsoft Corporation, with NumPy by the NumPy Developers and, on Windows, Microsoft's DirectML — the AI runtime add-on that runs the optional AI models, an optional download
- DeepFilterNet by Hendrik Schröter — speech enhancement for Audio Restoration, an optional download (MIT or Apache-2.0)
- libdvdread by VideoLAN — DVD reading
- FFmpeg — encoding and decoding
- Hybrid by Selur — the inspiration for this project
Full licence texts, copyright holders and every bundled component are listed in licenses/NOTICES.txt. If an attribution there is wrong, please open an issue and it will be fixed.
Pre-built binary sources
Components taken pre-built rather than compiled from source come unmodified from:
- yuygfgg/Macos_vapoursynth_plugins — ARM64 VapourSynth plugins for macOS
- Stefan-Olt/vs-plugin-build — cross-platform VapourSynth plugins (used for
tmedian) - evermeet.cx — static FFmpeg for Intel Macs
- ffmpeg.martin-riedl.de — static FFmpeg for Apple Silicon
- BtbN/FFmpeg-Builds — static FFmpeg for Windows and Linux
- python-build-standalone — the relocatable Python used on macOS and Linux
- python.org — the embeddable Python used on Windows
- PyPI — the plugins their authors publish as wheels: akarin, bwdif, dedot, descale, SangNom and vszip (and on Windows also BM3D and KNLMeansCL)
- The plugins' own GitHub releases — most Windows plugins, VapourSynth itself on Windows, and zsmooth
- styler00dollar/VapourSynth-RIFE-ncnn-Vulkan — the RIFE plugin
- AmusementClub/vs-mlrt — VSNCNN on Windows (macOS and Linux build it from source)
- Tencent/ncnn — the ncnn library VSNCNN is built against on macOS and Linux
- KhronosGroup/MoltenVK — Vulkan on macOS
- Homebrew — the
whisper-cppbuild used for the macOS subtitle add-on
The Intel Mac build compiles its support libraries from source to run on macOS 12, and is the one build without akarin, whose only macOS build needs macOS 14. Intel Macs don't need it: VapourSynth's own expression compiler covers them.