Skip to content

blur

Functions:

  • bilateral

    Applies a bilateral filter for edge-preserving and noise-reducing smoothing.

  • box_blur

    Applies a box blur to the input clip.

  • flux_smooth

    FluxSmoothT examines each pixel and compares it to the corresponding pixel in the previous and next frames.

  • gauss_blur

    Applies Gaussian blur to a clip, supporting spatial and temporal modes, and per-plane control.

  • guided_filter
  • median_blur

    Applies a median blur to the clip using spatial or temporal neighborhood.

  • min_blur

    Combines binomial (Gaussian-like) blur and median filtering for a balanced smoothing effect.

  • sbr

    A helper function for high-pass filtering a blur difference, inspired by an AviSynth script by Didée.

  • side_box_blur

Bilateral

Bilateral(bilateral_func: Callable[P, R])

Bases: Generic[P, R]

Class decorator that wraps the bilateral function and extends its functionality.

It is not meant to be used directly.

Classes:

  • Backend

    Enum specifying which backend implementation of the bilateral filter to use.

Methods:

Source code
444
445
def __init__(self, bilateral_func: Callable[P, R]) -> None:
    self._func = bilateral_func

Backend

Bases: CustomStrEnum

Enum specifying which backend implementation of the bilateral filter to use.

Methods:

  • Bilateral

    Applies the bilateral filter using the plugin associated with the selected backend.

Attributes:

  • CPU

    Uses vszip.Bilateral — a fast, CPU-based implementation written in Zig.

  • GPU

    Uses bilateralgpu.Bilateral — a CUDA-based GPU implementation.

  • GPU_RTC

    Uses bilateralgpu_rtc.Bilateral — a CUDA-based GPU implementation with runtime shader compilation.

CPU class-attribute instance-attribute

CPU = 'vszip'

Uses vszip.Bilateral — a fast, CPU-based implementation written in Zig.

GPU class-attribute instance-attribute

GPU = 'bilateralgpu'

Uses bilateralgpu.Bilateral — a CUDA-based GPU implementation.

GPU_RTC class-attribute instance-attribute

GPU_RTC = 'bilateralgpu_rtc'

Uses bilateralgpu_rtc.Bilateral — a CUDA-based GPU implementation with runtime shader compilation.

Bilateral

Bilateral(
    clip: VideoNode, *args: Any, **kwargs: Any
) -> ConstantFormatVideoNode

Applies the bilateral filter using the plugin associated with the selected backend.

Parameters:

  • clip
    (VideoNode) –

    Source clip.

  • *args
    (Any, default: () ) –

    Positional arguments passed to the selected plugin.

  • **kwargs
    (Any, default: {} ) –

    Keyword arguments passed to the selected plugin.

Returns:

  • ConstantFormatVideoNode

    Bilaterally filtered clip.

Source code
470
471
472
473
474
475
476
477
478
479
480
481
482
def Bilateral(self, clip: vs.VideoNode, *args: Any, **kwargs: Any) -> ConstantFormatVideoNode:  # noqa: N802
    """
    Applies the bilateral filter using the plugin associated with the selected backend.

    Args:
        clip: Source clip.
        *args: Positional arguments passed to the selected plugin.
        **kwargs: Keyword arguments passed to the selected plugin.

    Returns:
        Bilaterally filtered clip.
    """
    return getattr(clip, self.value).Bilateral(*args, **kwargs)

__call__

__call__(*args: args, **kwargs: kwargs) -> R
Source code
447
448
def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R:
    return self._func(*args, **kwargs)

GuidedFilter

GuidedFilter(guided_filter_func: Callable[P, R])

Bases: Generic[P, R]

Class decorator that wraps the guided_filter function and extends its functionality.

It is not meant to be used directly.

Classes:

Methods:

Source code
584
585
def __init__(self, guided_filter_func: Callable[P, R]) -> None:
    self._func = guided_filter_func

Mode

Bases: CustomIntEnum

Attributes:

  • GRADIENT

    Gradient Domain Guided Image Filter

  • ORIGINAL

    Original Guided Filter

  • WEIGHTED

    Weighted Guided Image Filter

GRADIENT class-attribute instance-attribute

GRADIENT = 2

Gradient Domain Guided Image Filter

ORIGINAL class-attribute instance-attribute

ORIGINAL = 0

Original Guided Filter

WEIGHTED class-attribute instance-attribute

WEIGHTED = 1

Weighted Guided Image Filter

__call__

__call__(*args: args, **kwargs: kwargs) -> R
Source code
587
588
def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R:
    return self._func(*args, **kwargs)

bilateral

bilateral(
    clip: VideoNode,
    ref: VideoNode | None = None,
    sigmaS: float | Sequence[float] | None = None,
    sigmaR: float | Sequence[float] | None = None,
    backend: Backend = CPU,
    **kwargs: Any
) -> ConstantFormatVideoNode

Applies a bilateral filter for edge-preserving and noise-reducing smoothing.

This filter replaces each pixel with a weighted average of nearby pixels based on both spatial distance and pixel intensity similarity. It can be used for joint (cross) bilateral filtering when a reference clip is given.

Example
blurred = bilateral(clip, ref, 3.0, 0.02, backend=bilateral.Backend.CPU)

For more details, see: - https://github.com/dnjulek/vapoursynth-zip/wiki/Bilateral - https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Bilateral - https://github.com/WolframRhodium/VapourSynth-BilateralGPU

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • ref

    (VideoNode | None, default: None ) –

    Optional reference clip for joint bilateral filtering.

  • sigmaS

    (float | Sequence[float] | None, default: None ) –

    Spatial sigma (controls the extent of spatial smoothing). Can be a float or per-plane list.

  • sigmaR

    (float | Sequence[float] | None, default: None ) –

    Range sigma (controls sensitivity to intensity differences). Can be a float or per-plane list.

  • backend

    (Backend, default: CPU ) –

    Backend implementation to use.

  • **kwargs

    (Any, default: {} ) –

    Additional arguments forwarded to the backend-specific implementation.

Returns:

  • ConstantFormatVideoNode

    Bilaterally filtered clip.

Source code
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
@Bilateral
def bilateral(
    clip: vs.VideoNode,
    ref: vs.VideoNode | None = None,
    sigmaS: float | Sequence[float] | None = None,  # noqa: N803
    sigmaR: float | Sequence[float] | None = None,  # noqa: N803
    backend: Bilateral.Backend = Bilateral.Backend.CPU,
    **kwargs: Any,
) -> ConstantFormatVideoNode:
    """
    Applies a bilateral filter for edge-preserving and noise-reducing smoothing.

    This filter replaces each pixel with a weighted average of nearby pixels based on both spatial distance
    and pixel intensity similarity.
    It can be used for joint (cross) bilateral filtering when a reference clip is given.

    Example:
        ```py
        blurred = bilateral(clip, ref, 3.0, 0.02, backend=bilateral.Backend.CPU)
        ```

    For more details, see:
        - https://github.com/dnjulek/vapoursynth-zip/wiki/Bilateral
        - https://github.com/HomeOfVapourSynthEvolution/VapourSynth-Bilateral
        - https://github.com/WolframRhodium/VapourSynth-BilateralGPU

    Args:
        clip: Source clip.
        ref: Optional reference clip for joint bilateral filtering.
        sigmaS: Spatial sigma (controls the extent of spatial smoothing). Can be a float or per-plane list.
        sigmaR: Range sigma (controls sensitivity to intensity differences). Can be a float or per-plane list.
        backend: Backend implementation to use.
        **kwargs: Additional arguments forwarded to the backend-specific implementation.

    Returns:
        Bilaterally filtered clip.
    """
    assert check_variable_format(clip, bilateral)

    if backend == Bilateral.Backend.CPU:
        bilateral_args = KwargsT(ref=ref, sigmaS=sigmaS, sigmaR=sigmaR, planes=normalize_planes(clip))
    else:
        bilateral_args = KwargsT(ref=ref, sigma_spatial=sigmaS, sigma_color=sigmaR)

    return backend.Bilateral(clip, **bilateral_args | kwargs)

box_blur

box_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    passes: int = 1,
    mode: OneDimConvModeT | TempConvModeT = HV,
    planes: PlanesT = None,
    **kwargs: Any
) -> ConstantFormatVideoNode

Applies a box blur to the input clip.

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • radius

    (int | Sequence[int], default: 1 ) –

    Blur radius (spatial or temporal) Can be a int or a list for per-plane control. Defaults to 1

  • passes

    (int, default: 1 ) –

    Number of times the blur is applied. Defaults to 1

  • mode

    (OneDimConvModeT | TempConvModeT, default: HV ) –

    Convolution mode (horizontal, vertical, both, or temporal). Defaults to HV.

  • planes

    (PlanesT, default: None ) –

    Planes to process. Defaults to all.

Raises:

  • CustomValueError

    If square convolution mode is specified, which is unsupported.

Returns:

  • ConstantFormatVideoNode

    Blurred clip.

Source code
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
def box_blur(
    clip: vs.VideoNode,
    radius: int | Sequence[int] = 1,
    passes: int = 1,
    mode: OneDimConvModeT | TempConvModeT = ConvMode.HV,
    planes: PlanesT = None,
    **kwargs: Any,
) -> ConstantFormatVideoNode:
    """
    Applies a box blur to the input clip.

    Args:
        clip: Source clip.
        radius: Blur radius (spatial or temporal) Can be a int or a list for per-plane control. Defaults to 1
        passes: Number of times the blur is applied. Defaults to 1
        mode: Convolution mode (horizontal, vertical, both, or temporal). Defaults to HV.
        planes: Planes to process. Defaults to all.

    Raises:
        CustomValueError: If square convolution mode is specified, which is unsupported.

    Returns:
        Blurred clip.
    """
    assert check_variable(clip, box_blur)

    if isinstance(radius, Sequence):
        return normalize_radius(clip, box_blur, radius, planes, passes=passes, mode=mode, **kwargs)

    if not radius:
        return clip

    if mode == ConvMode.TEMPORAL:
        return BlurMatrix.MEAN(radius, mode=mode)(clip, planes, passes=passes, **kwargs)

    if not TYPE_CHECKING and mode == ConvMode.SQUARE:
        raise CustomValueError("Invalid mode specified", box_blur, mode)

    box_args = (
        planes,
        radius,
        0 if mode == ConvMode.VERTICAL else passes,
        radius,
        0 if mode == ConvMode.HORIZONTAL else passes,
    )

    return clip.vszip.BoxBlur(*box_args)

flux_smooth

flux_smooth(
    clip: VideoNode,
    temporal_threshold: float | Sequence[float] = 7.0,
    spatial_threshold: float | Sequence[float] | None = None,
    planes: PlanesT = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode

FluxSmoothT examines each pixel and compares it to the corresponding pixel in the previous and next frames. Smoothing occurs if both the previous frame's value and the next frame's value are greater, or if both are less than the value in the current frame.

Smoothing is done by averaging the pixel from the current frame with the pixels from the previous and/or next frames, if they are within temporal_threshold.

FluxSmoothST does the same as FluxSmoothT, except the pixel's eight neighbours from the current frame are also included in the average, if they are within spatial_threshold.

The first and last rows and the first and last columns are not processed by FluxSmoothST.

Parameters:

  • clip

    (VideoNode) –

    Clip to process.

  • temporal_threshold

    (float | Sequence[float], default: 7.0 ) –

    Temporal neighbour pixels within this threshold from the current pixel are included in the average. Can be specified as an array, with values corresonding to each plane of the input clip. A negative value (such as -1) indicates that the plane should not be processed and will be copied from the input clip.

  • spatial_threshold

    (float | Sequence[float] | None, default: None ) –

    Spatial neighbour pixels within this threshold from the current pixel are included in the average. A negative value (such as -1) indicates that the plane should not be processed and will be copied from the input clip.

  • planes

    (PlanesT, default: None ) –

    Which planes to process. Default to all.

  • scalep

    (bool, default: True ) –

    Parameter scaling. If set to true, all threshold values will be automatically scaled from 8-bit range (0-255) to the corresponding range of the input clip's bit depth.

Returns:

  • ConstantFormatVideoNode

    Smoothed clip.

Source code
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
def flux_smooth(
    clip: vs.VideoNode,
    temporal_threshold: float | Sequence[float] = 7.0,
    spatial_threshold: float | Sequence[float] | None = None,
    planes: PlanesT = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode:
    """
    FluxSmoothT examines each pixel and compares it to the corresponding pixel in the previous and next frames.
    Smoothing occurs if both the previous frame's value and the next frame's value are greater,
    or if both are less than the value in the current frame.

    Smoothing is done by averaging the pixel from the current frame with the pixels from the previous
    and/or next frames, if they are within temporal_threshold.

    FluxSmoothST does the same as FluxSmoothT, except the pixel's eight neighbours from the current frame
    are also included in the average, if they are within spatial_threshold.

    The first and last rows and the first and last columns are not processed by FluxSmoothST.

    Args:
        clip: Clip to process.
        temporal_threshold: Temporal neighbour pixels within this threshold from the current pixel are included in the
            average. Can be specified as an array, with values corresonding to each plane of the input clip. A negative
            value (such as -1) indicates that the plane should not be processed and will be copied from the input clip.
        spatial_threshold: Spatial neighbour pixels within this threshold from the current pixel are included in the
            average. A negative value (such as -1) indicates that the plane should not be processed and will be copied
            from the input clip.
        planes: Which planes to process. Default to all.
        scalep: Parameter scaling. If set to true, all threshold values will be automatically scaled from 8-bit range
            (0-255) to the corresponding range of the input clip's bit depth.

    Returns:
        Smoothed clip.
    """

    assert check_variable_format(clip, flux_smooth)

    if spatial_threshold:
        return core.zsmooth.FluxSmoothST(clip, temporal_threshold, spatial_threshold, planes, scalep)

    return core.zsmooth.FluxSmoothT(clip, temporal_threshold, planes, scalep)

gauss_blur

gauss_blur(
    clip: VideoNode,
    sigma: float | Sequence[float] = 0.5,
    taps: int | None = None,
    mode: OneDimConvModeT | TempConvModeT = HV,
    planes: PlanesT = None,
    **kwargs: Any
) -> ConstantFormatVideoNode

Applies Gaussian blur to a clip, supporting spatial and temporal modes, and per-plane control.

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • sigma

    (float | Sequence[float], default: 0.5 ) –

    Standard deviation of the Gaussian kernel. Can be a float or a list for per-plane control.

  • taps

    (int | None, default: None ) –

    Number of taps in the kernel. Automatically determined if not specified.

  • mode

    (OneDimConvModeT | TempConvModeT, default: HV ) –

    Convolution mode (horizontal, vertical, both, or temporal). Defaults to HV.

  • planes

    (PlanesT, default: None ) –

    Planes to process. Defaults to all.

  • **kwargs

    (Any, default: {} ) –

    Additional arguments passed to the resizer or blur kernel. Specifying _fast=True enables fast approximation.

Raises:

  • CustomValueError

    If square convolution mode is specified, which is unsupported.

Returns:

  • ConstantFormatVideoNode

    Blurred clip.

Source code
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
def gauss_blur(
    clip: vs.VideoNode,
    sigma: float | Sequence[float] = 0.5,
    taps: int | None = None,
    mode: OneDimConvModeT | TempConvModeT = ConvMode.HV,
    planes: PlanesT = None,
    **kwargs: Any,
) -> ConstantFormatVideoNode:
    """
    Applies Gaussian blur to a clip, supporting spatial and temporal modes, and per-plane control.

    Args:
        clip: Source clip.
        sigma: Standard deviation of the Gaussian kernel. Can be a float or a list for per-plane control.
        taps: Number of taps in the kernel. Automatically determined if not specified.
        mode: Convolution mode (horizontal, vertical, both, or temporal). Defaults to HV.
        planes: Planes to process. Defaults to all.
        **kwargs: Additional arguments passed to the resizer or blur kernel. Specifying `_fast=True` enables fast
            approximation.

    Raises:
        CustomValueError: If square convolution mode is specified, which is unsupported.

    Returns:
        Blurred clip.
    """
    assert check_variable(clip, gauss_blur)

    planes = normalize_planes(clip, planes)

    if not TYPE_CHECKING and mode == ConvMode.SQUARE:
        raise CustomValueError("Invalid mode specified", gauss_blur, mode)

    if isinstance(sigma, Sequence):
        return normalize_radius(clip, gauss_blur, {"sigma": sigma}, planes, mode=mode)

    fast = kwargs.pop("_fast", False)

    sigma_constant = 0.9 if fast and not mode.is_temporal else sigma
    taps = BlurMatrix.GAUSS.get_taps(sigma_constant, taps)

    if not mode.is_temporal:

        def _resize2_blur(plane: ConstantFormatVideoNode, sigma: float, taps: int) -> ConstantFormatVideoNode:
            resize_kwargs = dict[str, Any]()

            # Downscale approximation can be used by specifying _fast=True
            # Has a big speed gain when taps is large
            if fast:
                wdown, hdown = plane.width, plane.height

                if ConvMode.VERTICAL in mode:
                    hdown = round(max(round(hdown / sigma), 2) / 2) * 2

                if ConvMode.HORIZONTAL in mode:
                    wdown = round(max(round(wdown / sigma), 2) / 2) * 2

                resize_kwargs.update(width=plane.width, height=plane.height)

                plane = core.resize.Bilinear(plane, wdown, hdown)
                sigma = sigma_constant
            else:
                resize_kwargs.update({f"force_{k}": k in mode for k in "hv"})

            return Gaussian(sigma, taps).scale(plane, **resize_kwargs | kwargs)  # type: ignore[return-value]

        if not {*range(clip.format.num_planes)} - {*planes}:
            return _resize2_blur(clip, sigma, taps)

        return join([_resize2_blur(p, sigma, taps) if i in planes else p for i, p in enumerate(split(clip))])

    kernel = BlurMatrix.GAUSS(taps, sigma=sigma, mode=mode, scale_value=1023)

    return kernel(clip, planes, **kwargs)

guided_filter

guided_filter(
    clip: VideoNode,
    guidance: VideoNode | None = None,
    radius: int | Sequence[int] | None = None,
    thr: float | Sequence[float] = 1 / 3,
    mode: Mode = GRADIENT,
    use_gauss: bool = False,
    planes: PlanesT = None,
    down_ratio: int = 0,
    downscaler: ScalerLike = Point,
    upscaler: ScalerLike = Bilinear,
) -> VideoNode
Source code
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
@GuidedFilter
def guided_filter(
    clip: vs.VideoNode,
    guidance: vs.VideoNode | None = None,
    radius: int | Sequence[int] | None = None,
    thr: float | Sequence[float] = 1 / 3,
    mode: GuidedFilter.Mode = GuidedFilter.Mode.GRADIENT,
    use_gauss: bool = False,
    planes: PlanesT = None,
    down_ratio: int = 0,
    downscaler: ScalerLike = Point,
    upscaler: ScalerLike = Bilinear,
) -> vs.VideoNode:
    assert check_variable(clip, guided_filter)

    planes = normalize_planes(clip, planes)

    downscaler = Scaler.ensure_obj(downscaler, guided_filter)
    upscaler = Scaler.ensure_obj(upscaler, guided_filter)

    width, height = clip.width, clip.height

    thr = normalize_seq(thr, clip.format.num_planes)

    size = normalize_seq(
        [220, 225, 225] if ColorRange.from_video(clip, func=guided_filter).is_full else 256, clip.format.num_planes
    )

    thr = [t / s for t, s in zip(thr, size)]

    if radius is None:
        radius = [
            round(max((w - 1280) / 160 + 12, (h - 720) / 90 + 12))
            for w, h in [get_plane_sizes(clip, i) for i in range(clip.format.num_planes)]
        ]

    check_ref_clip(clip, guidance)

    p, bits = expect_bits(clip, 32)
    guidance_clip = g = depth(guidance, 32) if guidance is not None else p

    radius = normalize_seq(radius, clip.format.num_planes)

    if down_ratio:
        down_w, down_h = cround(width / down_ratio), cround(height / down_ratio)

        p = downscaler.scale(p, down_w, down_h)
        g = downscaler.scale(g, down_w, down_h) if guidance is not None else p

        radius = [cround(rad / down_ratio) for rad in radius]

    blur_filter = (
        partial(gauss_blur, sigma=[rad / 2 * sqrt(2) for rad in radius], planes=planes)
        if use_gauss
        else partial(box_blur, radius=[rad + 1 for rad in radius], planes=planes)
    )

    blur_filter_corr = (
        partial(gauss_blur, sigma=1 / 2 * sqrt(2), planes=planes)
        if use_gauss
        else partial(box_blur, radius=2, planes=planes)
    )

    mean_p = blur_filter(p)
    mean_I = blur_filter(g) if guidance is not None else mean_p

    I_square = norm_expr(g, "x dup *", planes, func=guided_filter)
    corr_I = blur_filter(I_square)
    corr_Ip = blur_filter(norm_expr([g, p], "x y *", planes, func=guided_filter)) if guidance is not None else corr_I

    var_I = norm_expr([corr_I, mean_I], "x y dup * -", planes, func=guided_filter)
    cov_Ip = (
        norm_expr([corr_Ip, mean_I, mean_p], "x y z * -", planes, func=guided_filter) if guidance is not None else var_I
    )

    if mode is GuidedFilter.Mode.ORIGINAL:
        a = norm_expr([cov_Ip, var_I], "x y {thr} + /", planes, thr=thr, func=guided_filter)
    else:
        if set(radius) == {1}:
            var_I_1 = var_I
        else:
            mean_I_1 = blur_filter_corr(g)
            corr_I_1 = blur_filter_corr(I_square)
            var_I_1 = norm_expr([corr_I_1, mean_I_1], "x y dup * -", planes, func=guided_filter)

        if mode is GuidedFilter.Mode.WEIGHTED:
            weight_in = var_I_1
        else:
            weight_in = norm_expr([var_I, var_I_1], "x y * sqrt", planes, func=guided_filter)

        denominator = norm_expr([weight_in], "1 x {eps} + /", planes, eps=1e-06, func=guided_filter)

        denominator = denominator.std.PlaneStats(None, 0)

        weight = norm_expr([weight_in, denominator], "x 1e-06 + y.PlaneStatsAverage *", planes, func=guided_filter)

        if mode is GuidedFilter.Mode.WEIGHTED:
            a = norm_expr([cov_Ip, var_I, weight], "x y {thr} z / + /", planes, thr=thr, func=guided_filter)
        else:
            weight_in = weight_in.std.PlaneStats(None, 0)

            a = norm_expr(
                [cov_Ip, weight_in, weight, var_I],
                "x {thr} 1 1 1 -4 y.PlaneStatsMin y.PlaneStatsAverage 1e-6 - - / "
                "y y.PlaneStatsAverage - * exp + / - * z / + a {thr} z / + /",
                planes,
                thr=thr,
            )

    b = norm_expr([mean_p, a, mean_I], "x y z * -", planes, func=guided_filter)

    mean_a, mean_b = blur_filter(a), blur_filter(b)

    if down_ratio:
        mean_a = upscaler.scale(mean_a, width, height)
        mean_b = upscaler.scale(mean_b, width, height)

    q = norm_expr([mean_a, guidance_clip, mean_b], "x y * z +", planes, func=guided_filter)

    return depth(q, bits)

median_blur

median_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: SpatialConvModeT = SQUARE,
    planes: PlanesT = None,
) -> ConstantFormatVideoNode
median_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: Literal[SQUARE] = ...,
    planes: PlanesT = None,
    smart: Literal[True] = ...,
    threshold: float | Sequence[float] | None = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode
median_blur(
    clip: VideoNode,
    radius: int = 1,
    mode: Literal[TEMPORAL] = ...,
    planes: PlanesT = None,
) -> ConstantFormatVideoNode
median_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: ConvMode = SQUARE,
    planes: PlanesT = None,
    smart: bool = False,
    threshold: float | Sequence[float] | None = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode
median_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: ConvMode = SQUARE,
    planes: PlanesT = None,
    smart: bool = False,
    threshold: float | Sequence[float] | None = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode

Applies a median blur to the clip using spatial or temporal neighborhood.

  • In temporal mode, each pixel is replaced by the median across multiple frames.
  • In spatial modes, each pixel is replaced with the median of its 2D neighborhood.

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • radius

    (int | Sequence[int], default: 1 ) –

    Blur radius per plane (list) or uniform radius (int). Only int is allowed in temporal mode.

  • mode

    (ConvMode, default: SQUARE ) –

    Convolution mode. Defaults to SQUARE.

  • planes

    (PlanesT, default: None ) –

    Planes to process. Defaults to all.

  • smart

    (bool, default: False ) –

    Enable Smart Median by zsmooth, thresholded based on a modified form of variance.

  • threshold

    (float | Sequence[float] | None, default: None ) –

    The variance threshold when smart=True. Pixels with a variance under the threshold are smoothed, and over the threshold are returned as is.

  • scalep

    (bool, default: True ) –

    Parameter scaling when smart=True. If True, all threshold values will be automatically scaled from 8-bit range (0-255) to the corresponding range of the input clip's bit depth.

Raises:

  • CustomValueError

    If a list is passed for radius in temporal mode, which is unsupported or if smart=True and mode != ConvMode.SQUARE.

Returns:

  • ConstantFormatVideoNode

    Median-blurred video clip.

Source code
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
def median_blur(
    clip: vs.VideoNode,
    radius: int | Sequence[int] = 1,
    mode: ConvMode = ConvMode.SQUARE,
    planes: PlanesT = None,
    smart: bool = False,
    threshold: float | Sequence[float] | None = None,
    scalep: bool = True,
) -> ConstantFormatVideoNode:
    """
    Applies a median blur to the clip using spatial or temporal neighborhood.

    - In temporal mode, each pixel is replaced by the median across multiple frames.
    - In spatial modes, each pixel is replaced with the median of its 2D neighborhood.

    Args:
        clip: Source clip.
        radius: Blur radius per plane (list) or uniform radius (int). Only int is allowed in temporal mode.
        mode: Convolution mode. Defaults to SQUARE.
        planes: Planes to process. Defaults to all.
        smart: Enable [Smart Median by zsmooth](https://github.com/adworacz/zsmooth?tab=readme-ov-file#smart-median),
            thresholded based on a modified form of variance.
        threshold: The variance threshold when ``smart=True``. Pixels with a variance under the threshold are smoothed,
            and over the threshold are returned as is.
        scalep: Parameter scaling when ``smart=True``. If True, all threshold values will be automatically scaled
            from 8-bit range (0-255) to the corresponding range of the input clip's bit depth.

    Raises:
        CustomValueError: If a list is passed for radius in temporal mode, which is unsupported
            or if smart=True and mode != ConvMode.SQUARE.

    Returns:
        Median-blurred video clip.
    """
    assert check_variable(clip, median_blur)

    if mode == ConvMode.TEMPORAL:
        if isinstance(radius, int):
            return clip.zsmooth.TemporalMedian(radius, planes)

        raise CustomValueError("A list of radius isn't supported for ConvMode.TEMPORAL!", median_blur, radius)

    radius = normalize_seq(radius, clip.format.num_planes)

    if smart:
        if mode == ConvMode.SQUARE:
            return core.zsmooth.SmartMedian(clip, radius, threshold, scalep, planes)

        raise CustomValueError("When using SmartMedian, mode should be ConvMode.SQUARE!", median_blur, mode)

    if mode == ConvMode.SQUARE and max(radius) <= 3:
        return core.zsmooth.Median(clip, radius, planes)

    if mode == ConvMode.VERTICAL and max(radius) <= 1:
        return vertical_cleaner(clip, radius, planes)

    expr_plane = list[list[str]]()

    for r in radius:
        expr_passes = list[str]()

        for mat in ExprOp.matrix("x", r, mode, [(0, 0)]):
            rb = len(mat) + 1
            st = rb - 1
            sp = rb // 2 - 1
            dp = st - 2

            expr_passes.append(f"{mat} sort{st} swap{sp} min! swap{sp} max! drop{dp} x min@ max@ clip")

        expr_plane.append(expr_passes)

    for e in zip(*expr_plane):
        clip = norm_expr(clip, e, planes, func=median_blur)

    return clip

min_blur

min_blur(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: tuple[ConvMode, ConvMode] = (HV, SQUARE),
    planes: PlanesT = None,
    **kwargs: Any
) -> ConstantFormatVideoNode

Combines binomial (Gaussian-like) blur and median filtering for a balanced smoothing effect.

This filter blends the input clip with both a binomial blur and a median blur to achieve a "best of both worlds" result — combining the edge-preserving nature of median filtering with the smoothness of Gaussian blur. The effect is somewhat reminiscent of a bilateral filter.

Original concept: http://avisynth.nl/index.php/MinBlur

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • radius

    (int | Sequence[int], default: 1 ) –

    Radius of blur to apply. Can be a single int or a list for per-plane control.

  • mode

    (tuple[ConvMode, ConvMode], default: (HV, SQUARE) ) –

    A tuple of two convolution modes: - First element: mode for binomial blur. - Second element: mode for median blur. Defaults to (HV, SQUARE).

  • planes

    (PlanesT, default: None ) –

    Planes to process. Defaults to all.

  • **kwargs

    (Any, default: {} ) –

    Additional arguments passed to the binomial blur.

Returns:

  • ConstantFormatVideoNode

    Clip with MinBlur applied.

Source code
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
def min_blur(
    clip: vs.VideoNode,
    radius: int | Sequence[int] = 1,
    mode: tuple[ConvMode, ConvMode] = (ConvMode.HV, ConvMode.SQUARE),
    planes: PlanesT = None,
    **kwargs: Any,
) -> ConstantFormatVideoNode:
    """
    Combines binomial (Gaussian-like) blur and median filtering for a balanced smoothing effect.

    This filter blends the input clip with both a binomial blur and a median blur to achieve
    a "best of both worlds" result — combining the edge-preserving nature of median filtering
    with the smoothness of Gaussian blur. The effect is somewhat reminiscent of a bilateral filter.

    Original concept: http://avisynth.nl/index.php/MinBlur

    Args:
        clip: Source clip.
        radius: Radius of blur to apply. Can be a single int or a list for per-plane control.
        mode: A tuple of two convolution modes:
               - First element: mode for binomial blur.
               - Second element: mode for median blur.
            Defaults to (HV, SQUARE).
        planes: Planes to process. Defaults to all.
        **kwargs: Additional arguments passed to the binomial blur.

    Returns:
        Clip with MinBlur applied.
    """
    assert check_variable(clip, min_blur)

    planes = normalize_planes(clip, planes)

    if isinstance(radius, Sequence):
        return normalize_radius(clip, min_blur, radius, planes)

    mode_blur, mode_median = normalize_seq(mode, 2)

    blurred = BlurMatrix.BINOMIAL(radius=radius, mode=mode_blur)(clip, planes=planes, **kwargs)
    median = median_blur(clip, radius, mode_median, planes=planes)

    return MeanMode.MEDIAN([clip, blurred, median], planes=planes)

sbr

sbr(
    clip: VideoNode,
    radius: int | Sequence[int] = 1,
    mode: ConvMode = HV,
    blur: _SbrBlurT | VideoNode = BINOMIAL,
    blur_diff: _SbrBlurT = BINOMIAL,
    planes: PlanesT = None,
    *,
    func: FuncExceptT | None = None,
    **kwargs: Any
) -> ConstantFormatVideoNode

A helper function for high-pass filtering a blur difference, inspired by an AviSynth script by Didée. https://forum.doom9.org/showthread.php?p=1584186#post1584186

Parameters:

  • clip

    (VideoNode) –

    Source clip.

  • radius

    (int | Sequence[int], default: 1 ) –

    Specifies the size of the blur kernels if blur or blur_diff is a BlurMatrix enum. Default to 1.

  • mode

    (ConvMode, default: HV ) –

    Specifies the convolution mode. Defaults to horizontal + vertical.

  • blur

    (_SbrBlurT | VideoNode, default: BINOMIAL ) –

    Blur kernel to apply to the original clip. Defaults to binomial.

  • blur_diff

    (_SbrBlurT, default: BINOMIAL ) –

    Blur kernel to apply to the difference clip. Defaults to binomial.

  • planes

    (PlanesT, default: None ) –

    Which planes to process. Defaults to all.

  • **kwargs

    (Any, default: {} ) –

    Additional arguments passed to blur kernel call.

Returns:

  • ConstantFormatVideoNode

    Sbr'd clip.

Source code
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
def sbr(
    clip: vs.VideoNode,
    radius: int | Sequence[int] = 1,
    mode: ConvMode = ConvMode.HV,
    blur: _SbrBlurT | vs.VideoNode = BlurMatrix.BINOMIAL,
    blur_diff: _SbrBlurT = BlurMatrix.BINOMIAL,
    planes: PlanesT = None,
    *,
    func: FuncExceptT | None = None,
    **kwargs: Any,
) -> ConstantFormatVideoNode:
    """
    A helper function for high-pass filtering a blur difference, inspired by an AviSynth script by Didée.
    `https://forum.doom9.org/showthread.php?p=1584186#post1584186`

    Args:
        clip: Source clip.
        radius: Specifies the size of the blur kernels if `blur` or `blur_diff` is a BlurMatrix enum. Default to 1.
        mode: Specifies the convolution mode. Defaults to horizontal + vertical.
        blur: Blur kernel to apply to the original clip. Defaults to binomial.
        blur_diff: Blur kernel to apply to the difference clip. Defaults to binomial.
        planes: Which planes to process. Defaults to all.
        **kwargs: Additional arguments passed to blur kernel call.

    Returns:
        Sbr'd clip.
    """
    func = func or sbr

    if isinstance(radius, Sequence):
        return normalize_radius(clip, min_blur, list(radius), planes)

    def _apply_blur(clip: ConstantFormatVideoNode, blur: _SbrBlurT | vs.VideoNode) -> ConstantFormatVideoNode:
        if isinstance(blur, Sequence):
            return BlurMatrix.custom(blur, mode)(clip, planes, **kwargs)

        if isinstance(blur, BlurMatrix):
            return blur(taps=radius, mode=mode)(clip, planes, **kwargs)

        blurred = blur(clip) if callable(blur) else blur

        assert check_variable_format(blurred, func)

        return blurred

    assert check_variable(clip, func)

    planes = normalize_planes(clip, planes)

    blurred = _apply_blur(clip, blur)

    diff = clip.std.MakeDiff(blurred, planes=planes)
    blurred_diff = _apply_blur(diff, blur_diff)

    return norm_expr(
        [clip, diff, blurred_diff],
        "y neutral - D1! y z - D2! D1@ D2@ xor x x D1@ abs D2@ abs < D1@ D2@ ? - ?",
        planes=planes,
        func=func,
    )

side_box_blur

side_box_blur(
    clip: VideoNode, radius: int = 1, planes: PlanesT = None
) -> ConstantFormatVideoNode
Source code
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
def side_box_blur(clip: vs.VideoNode, radius: int = 1, planes: PlanesT = None) -> ConstantFormatVideoNode:
    assert check_variable_format(clip, side_box_blur)

    half_kernel = [(1 if i <= 0 else 0) for i in range(-radius, radius + 1)]

    vrt_filters = [
        partial(core.std.Convolution, matrix=half_kernel, planes=planes, mode=ConvMode.VERTICAL),
        partial(core.std.Convolution, matrix=half_kernel[::-1], planes=planes, mode=ConvMode.VERTICAL),
        partial(box_blur, radius=radius, mode=ConvMode.VERTICAL, planes=planes),
    ]

    hrz_filters = [
        partial(core.std.Convolution, matrix=half_kernel, planes=planes, mode=ConvMode.HORIZONTAL),
        partial(core.std.Convolution, matrix=half_kernel[::-1], planes=planes, mode=ConvMode.HORIZONTAL),
        partial(box_blur, radius=radius, mode=ConvMode.HORIZONTAL, planes=planes),
    ]

    vrt_intermediates = (vrt_flt(clip) for vrt_flt in vrt_filters)
    intermediates = (
        hrz_flt(vrt_intermediate)
        for i, vrt_intermediate in enumerate(vrt_intermediates)
        for j, hrz_flt in enumerate(hrz_filters)
        if not i == j == 2
    )

    return reduce(lambda x, y: norm_expr([x, y, clip], "x z - abs y z - abs < x y ?", planes=planes), intermediates)