Vapoursynth QTGMC
QTGMC is a slow, very high quality motion-compensated bobber. It interpolates every field, removes bob shimmer with motion-compensated temporal smoothing, resharpens, and can denoise or keep grain. QTempGaussMC is a different filter with its own page: QTempGaussMC.
Filtering → (De-)Interlace/Telecine → Deinterlacer

What it does
QTGMC first bobs the clip (NNEDI3 by default) so that every field becomes a frame. A motion-compensated temporal blur then removes the up-and-down flicker of bobbed lines, and the result is resharpened and limited against the source. Optional stages denoise or keep the grain, and source matching and lossless modes bring the result closer to the original fields. The native output has one frame per field. To repair progressive material with the same filter, see QTGMC (denoise).
Settings
Main page
| Control | Parameter | Range / values | Hybrid default | Meaning |
|---|---|---|---|---|
| Preset | Preset | Draft, Ultra Fast, Super Fast, Very Fast, Faster, Fast, Medium, Slow, Slower, Very Slow, Placebo, Custom, Fast - grain | Fast | Speed/quality preset; slower presets search motion more thoroughly and smooth more carefully. Custom opens the custom page; "Fast - grain" is Fast with a fixed set of settings that keep the grain (noise is restored, low sharpness, lossless after temporal smoothing). |
| Bob | – | on/off | off | On: keep every frame (double rate); off: keep one frame of each pair. |
| Select | – | Even, Odd | Even | Which frame of each pair is kept when Bob is off. |
| Pad mod16 | – | on/off | off | Pad the frame to a multiple of 16 before QTGMC (otherwise to the minimum the format needs). |
| Show settings | ShowSettings | on/off | off | Prints the settings onto the frame. |
| Show noise | ShowNoise | on/off | off | Shows the extracted noise instead of the picture. |
| Tools | – | mv, expr, rg, median, nnedi3, fft3d, dfttest, nlmeans, grain families | – | Implementations of the plugins QTGMC uses; GPU implementations also get a device choice. |
Custom page
Only used with Preset Custom.
| Control | Parameter | Range / values | Hybrid default | Meaning |
|---|---|---|---|---|
| Preset | Preset | Draft … Placebo | Slower | Base preset of the custom settings; changing it resets Final temporal smoothing. |
| Final temporal smoothing | TR2 | 0 – 3 | 1 | Radius of the final temporal smoothing; more is more stable and softer. |
| Sharpness | Sharpness | 0 – 1.9 | 1.0 (0.2 with source matching) | Amount of resharpening; written only when the check box is ticked. |
| Source matching | SourceMatch | off, basic, refined, 2xrefined | off | Corrects the result back towards the original fields; refined modes are slower and more accurate. |
| Match preset / Match preset 2 | MatchPreset, MatchPreset2 | presets not slower than the base preset | default | Presets of the source matching passes; default lets QTGMC choose. |
| Match enhance | MatchEnhance | 0 – 1000 | 0.5 | How strongly the refined source matching adds back detail. |
| Lossless | Lossless | off, after temporal smooth, before resharpening | off | Puts the original field lines back into the output: after the final smoothing (truly lossless, may show shimmer or combing), or before resharpening (more stable, not exactly lossless). |
| EZDenoise / EZKeepGrain | EZDenoise, EZKeepGrain | 0 – 100 | off | Denoise, or keep grain, by this amount; the two exclude each other. |
| Noise preset | NoisePreset | Slower … Faster | Fast | Speed of the noise processing; only with EZDenoise or EZKeepGrain. |
| Denoiser | Denoiser | auto, FFT3D, dfttest, KNLMeansCL | auto | Denoiser used by the noise processing; auto lets QTGMC choose. |
| EdiExt | EdiExt | BWDif; with the torch add-on (Windows only) also DDD, DeF, MFDIN(old), MFDIN(old)-TRT | off | Use an external interpolated clip instead of QTGMC's own interpolation. |
| Addition | – | text | empty | Further QTGMC arguments, appended to the call as written. |
Presets
A preset only fills in the values that are not given; every value set explicitly in Hybrid (on the Custom page or in Addition) overrides the preset. With any preset other than Custom and "Fast - grain", Hybrid passes only the preset, so the values below apply unchanged.
Preset settings
Values for Tuning "None" (QTGMC's default).
| Setting | Parameter | Placebo | Very Slow | Slower | Slow | Medium | Fast | Faster | Very Fast | Super Fast | Ultra Fast | Draft |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Motion search smoothing radius | TR0 | 2 | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 1 | 1 | 0 |
| Shimmer removal smoothing radius | TR1 | 2 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Final temporal smoothing (Custom page) | TR2 | 3 | 2 | 1 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 |
| Repair of the motion search clip | Rep0 | 4 | 4 | 4 | 4 | 3 | 3 | 0 | 0 | 0 | 0 | 0 |
| Repair of the final output | Rep2 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 3 | 3 | 0 |
| Interpolation | EdiMode | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | Bwdif | Bob |
| NNEDI3 predictor area | NNSize | 1 (16x6) | 1 (16x6) | 1 (16x6) | 1 (16x6) | 5 (16x4) | 5 (16x4) | 4 (8x4) | 4 (8x4) | 4 (8x4) | 4 (8x4) | 4 (8x4) |
| NNEDI3 neurons | NNeurons | 2 (64) | 2 (64) | 1 (32) | 1 (32) | 1 (32) | 0 (16) | 0 (16) | 0 (16) | 0 (16) | 0 (16) | 0 (16) |
| EEDI3 search distance (only with an EEDI3 EdiMode) | EdiMaxD | 12 | 10 | 8 | 7 | 7 | 6 | 6 | 5 | 4 | 4 | 4 |
| Resharpening mode | SMode | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 0 (off) |
| Sharpness (Custom page) | Sharpness | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 0.0 |
| Sharpness limiting | SLMode | 2 (temporal) | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 0 (off) | 0 | 0 |
| Sharpness limiting radius | SLRad | 3 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Back blend | Sbb | 3 (both) | 1 (before) | 1 | 0 (off) | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Motion search prefilter | SrchClipPP | 3 (Gauss + edge soften) | 3 | 3 | 3 | 3 | 2 (Gauss) | 2 | 2 | 1 (simple blur) | 1 | 0 (none) |
| Sub-pixel accuracy | SubPel | 2 (1/2) | 2 | 2 | 2 | 1 (full) | 1 | 1 | 1 | 1 | 1 | 1 |
| Block size | BlockSize | 16 | 16 | 16 | 16 | 16 | 16 | 32 | 32 | 32 | 32 | 32 |
| Block overlap | Overlap | 8 | 8 | 8 | 8 | 8 | 8 | 16 | 8 | 8 | 8 | 8 |
| Search method | Search | 5 (UMH) | 4 (hexagon) | 4 | 4 | 4 | 4 | 4 | 4 | 0 (one time) | 0 | 0 |
| Search range | SearchParam | 2 | 2 | 2 | 2 | 2 | 2 | 2 | 1 | 1 | 1 | 1 |
| Search range at sub-pixel level | PelSearch | 2 | 2 | 2 | 2 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Chroma in motion analysis | ChromaMotion | True | True | True | False | False | False | False | False | False | False | False |
| Vector coherence (from BlockSize) | Lambda | 400 | 400 | 400 | 400 | 400 | 400 | 1600 | 1600 | 1600 | 1600 | 1600 |
| Precise algorithms | Precise | True | True | False | False | False | False | False | False | False | False | False |
| Noise processing (without EZDenoise/EZKeepGrain) | NoiseProcess | 2 (keep grain) | 2 | 0 (off) | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
| Match preset (default) | MatchPreset | Slow | Medium | Fast | Faster | Very Fast | Super Fast | Ultra Fast | Ultra Fast | Ultra Fast | Ultra Fast | – |
| Match preset 2 (default) | MatchPreset2 | Fast | Faster | Very Fast | Super Fast | Ultra Fast | Ultra Fast | Ultra Fast | Ultra Fast | Ultra Fast | Ultra Fast | – |
- With Source matching on, Sharpness defaults to 0.2 and SLMode to 0. Draft does not support source matching.
- Tuning "DV-HD" (only through Addition) uses BlockSize 32 for every preset; Overlap is then 16 from Placebo to Faster.
Match preset settings
Interpolation values used by the source matching passes (Custom page, Source matching). The basic pass uses Match preset, the refined passes use Match preset 2; Draft is not available here.
| Setting | Parameter | Placebo | Very Slow | Slower | Slow | Medium | Fast | Faster | Very Fast | Super Fast | Ultra Fast |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Interpolation of the basic pass | MatchEdi | same as EdiMode of the main preset | Bwdif | ||||||||
| Interpolation of the refined passes | MatchEdi2 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | NNEDI3 | Bob |
| NNEDI3 predictor area | NNSize | 1 (16x6) | 1 (16x6) | 1 (16x6) | 1 (16x6) | 5 (16x4) | 5 (16x4) | 4 (8x4) | 4 (8x4) | 4 (8x4) | 4 (8x4) |
| NNEDI3 neurons | NNeurons | 2 (64) | 2 (64) | 1 (32) | 1 (32) | 1 (32) | 0 (16) | 0 (16) | 0 (16) | 0 (16) | 0 (16) |
| EEDI3 search distance | EdiMaxD | 12 | 10 | 8 | 7 | 7 | 6 | 6 | 5 | 4 | 4 |
Noise preset settings
Used with EZDenoise or EZKeepGrain (Custom page, Noise preset), and with the noise processing that Placebo and Very Slow switch on (then always Fast). Denoiser auto takes the value of this table.
| Setting | Parameter | Slower | Slow | Medium | Fast | Faster |
|---|---|---|---|---|---|---|
| Denoiser (Custom page) | Denoiser | dfttest | dfttest | dfttest | FFT3D | FFT3D |
| Motion-compensated denoising | DenoiseMC | True | True | False | False | False |
| Noise analysis radius | NoiseTR | 2 | 1 | 1 | 1 | 0 |
| Deinterlacing of the noise | NoiseDeint | Generate | Bob | DoubleWeave | DoubleWeave | DoubleWeave |
| Noise stabilization | StabilizeNoise | True | True | True | False | False |
EZDenoise and EZKeepGrain
| Parameter | neither (Placebo, Very Slow) | EZDenoise = d | EZKeepGrain = k |
|---|---|---|---|
| NoiseProcess | 2 | 1 (denoise) | 2 (keep grain) |
| Sigma | 2.0 | d | 4 × k |
| GrainRestore | 0.3 | 0.0 | 0.3 × √k |
| NoiseRestore | 0.1 | 0.0 | 0.1 × √k |
With the other presets and neither value set, noise processing is off.
Hybrid profile "Fast - grain"
"Fast - grain" runs Preset Fast and sets these values on top of it; everything not listed keeps the Fast value.
| Setting | Parameter | Fast | Fast - grain |
|---|---|---|---|
| Lossless | Lossless | 0 (off) | 1 (after temporal smooth) |
| Sharpness | Sharpness | 1.0 | 0.3 |
| Motion search smoothing radius | TR0 | 2 | 2 |
| Shimmer removal smoothing radius | TR1 | 1 | 2 |
| Repair of the motion search clip | Rep0 | 3 | 1 |
| Repair of the initial output | Rep1 | 0 | 0 |
| Repair of the final output | Rep2 | 4 | 4 |
| Block matching with DCT/SATD | DCT | 0 | 5 |
| Scene change threshold 1 | ThSCD1 | 180 | 300 |
| Scene change threshold 2 | ThSCD2 | 98 | 110 |
| Back blend | Sbb | 0 | 0 |
| Noise processing | NoiseProcess | 0 (off) | 2 (keep grain) |
| Grain restored before final smoothing | GrainRestore | 0.0 | 0.0 |
| Noise restored after final smoothing | NoiseRestore | 0.0 | 0.4 |
| Noise stabilization | StabilizeNoise | False | False |
| Noise analysis radius | NoiseTR | 0 (noise processing off) | 0 |
| Deinterlacing of the noise | NoiseDeint | DoubleWeave | Bob |
Example
Default settings on an 8-bit YUV 4:2:0 interlaced source (704x576, 25 fps, top field first):
# Deinterlacing using QTGMC
clip = qtgmc.QTGMC(clip, Preset="Fast", TFF=True, tools={'mv': 'mvutensils', 'expr': 'akarin', 'nnedi3': 'znedi3', 'eedi3': 'vszip', 'dfttest': 'neo_fft', 'nlmeans': 'knlm', 'bm3d': 'bm3d', 'rg': 'zsmooth', 'median': 'zsmooth', 'fft3d': 'neo_fft', 'grain': 'noise'}) # new fps: 50
# Making sure content is perceived as frame based
clip = core.std.SetFrameProps(clip=clip, _FieldBased=vs.FIELD_PROGRESSIVE) # scan type: progressive
clip = clip[::2] # selecting previously even frames, new fps: 25
How Hybrid calls it
- Field order: bottom field first sources run with TFF off, every other interlaced scan type with TFF on (detected scan type or Overwrite input scan type).
- Output frame rate: QTGMC always doubles the frame rate; without Bob Hybrid keeps every second frame (Even or Odd), so the source rate stays.
- Afterwards the clip counts as progressive. No decimation.
- Works on 8–16-bit YUV (not 4:1:1) and 8-bit gray; 4:2:0 needs a frame size that is a multiple of 4.
- With Custom, Hybrid always writes the base preset, Final temporal smoothing, Source matching and Lossless; Show settings and Show noise then only appear in the preview.
- QTGMC is also used as the helper deinterlacer of TIVTC, VIVTC, TFMBob and Fix combed with deinterlacer.
Background
QTGMC (by -Vit-) is a rewrite of Didée's TempGaussMC for AviSynth; the VapourSynth version descends from the havsfunc port (QTGMC 3.33). Hybrid ships its own maintained copy of that script, in which the implementation of each plugin can be chosen. Presets are not case sensitive; every other argument overrides what the preset sets. QTempGaussMC from vs-jetpack is a separate rewrite with different parameters and scales; settings cannot be carried over between the two.
Sources
- havsfunc (archived)
- Hybrid's tooltips