PARAMETERS.MD

Anime Scripter Parameters

Table of Contents

• Anime Scripter Parameters
  • Table of Contents
  • 1. Overview
  • 2. Command-Line Arguments
    • Offline Mode Options
  • 3. Input and Output
    • Examples
      • Multiple Input Files
  • Processing Options
    • Upscaling
      • Upscale Methods
    • Interpolation
      • Interpolate Methods
    • Scene Change Detection
      • Scene Change Methods
    • Deduplication
      • Dedup Methods
    • Restoring
      • Restore Methods
    • Sharpening
    • Segmentation
      • Segment Methods
    • Auto Cut Clips Detection
    • Depth Map Generation
      • Depth Methods
      • Depth Quality selection
    • Resizing
      • Resize Methods
  • Preset Arguments
  • Encoding Settings
• Encode Methods
  • Standard Encoders
  • Hardware Encoders
  • Other Formats
  • High Quality Presets
  • Performance Settings
  • Usage Examples

1. Overview

The Anime Scripter is a powerful tool for enhancing and manipulating videos with a wide range of options. It supports functionalities such as interpolation, upscaling, deduplication, denoising, sharpening, scene change detection, depth map generation, and more. Users can customize the processing pipeline to meet their specific requirements using a comprehensive set of parameters.

2. Command-Line Arguments

Argument	Type	Default	Description
--version	flag	-	Outputs the script version
--benchmark	flag	-	Enable benchmarking (no video output, performance testing only)
--offline	str	"none"	Download models based on user selection
--ae	flag	False	Indicates if the script is run from the After Effects interface
--preview	flag	False	Open localhost and request a preview frame from TAS
--realtime	flag	False	Open FFPLAY and watch the encoded video in real time

Offline Mode Options

• "none": Default state, does not download anything
• "all": Downloads all available models
• Single Option (e.g., "rife4.20"): Downloads the specified model
• Multiple Options (e.g., "rife4.20 rife4.18 shufflecugan-tensorrt"): Downloads the specified models

3. Input and Output

Argument	Type	Default	Description
--input	str	-	Absolute path of the input video(s). Multiple paths should be separated by semicolons (;).
--output	str	-	Output string or absolute path for the processed video

• NOTE: --input supports (".mp4", ".mkv", ".mov", ".avi", ".webm") containers as well as .txt, inputs in .txt should be separated by a new line.
• In order to download and process anime episodes, make sure to use the keyword "anime".

Examples

Multiple Input Files

--input "G:\TheAnimeScripter\input.mp4;G:\TheAnimeScripter\test.mp4"

# Or if the input is a txt, this is how the process should look like
--input "G:\TheAnimeScripter\test.txt"

# Contents of TXT:
"G:\TheAnimeScripter\test1.mp4" # separate by a new line here
"G:\TheAnimeScripter\test2.mp4"

Processing Options

Upscaling

Argument	Type	Default	Description
--upscale	flag	-	Enable upscaling
--upscale_factor	int	2	Factor for upscaling
--upscale_method	str	"ShuffleCugan"	Upscaling method
--custom_model	str	""	Path to a custom model file (.pth or .onnx)
--upscale_skip	flag	-	Skip processing duplicates for faster perceived upscaling

Upscale Methods

• "shufflecugan" / "shufflecugan-tensorrt"
• "compact" / "ultracompact" / "superultracompact" (with -directml or -tensorrt variants)
• "span" / "span-directml" / "span-tensorrt"
• "open-proteus" / "open-proteus-tensorrt" / "open-proteus-directml"
• "aniscale2" / "aniscale2-tensorrt" / "aniscale2-directml"
• "rtmosr-tensorrt" / "rtmosr-directml"

Interpolation

Argument	Type	Default	Description
--interpolate	flag	-	Enable interpolation
--interpolate_factor	int	2	Factor for interpolation
--interpolate_method	str	"rife"	Interpolation method
--ensemble	flag	-	Activate Ensemble for higher quality Rife outputs
--interpolate_skip	flag	-	Enable duplicate frame skipping in order to improve performance
--dynamic_scale	flag	-	Enable Dynamic Flow Scaling, which can enhance output quality with a significant performance trade-off. CUDA support only.

Interpolate Methods

• "rife4.6" / "rife4.6-ncnn" / "rife4.6-tensorrt"
• "rife4.15-lite" / "rife4.15-lite-ncnn" / "rife4.15-lite-tensorrt"
• "rife4.16-lite" / "rife4.16-lite-ncnn"
• "rife4.17" / "rife4.17-tensorrt" / "rife4.17-ncnn"
• "rife4.18" / "rife4.18-tensorrt" / "rife4.18-ncnn"
• "rife4.20" / "rife4.20-tensorrt" / "rife4.20-ncnn"
• "rife4.21" / "rife4.21-tensorrt" / "rife4.21-ncnn"
• "rife4.22" / "rife4.22-tensorrt" / "rife4.22-ncnn"
• "rife4.22-lite" / "rife4.22-lite-tensorrt" / "rife4.22-lite-ncnn"
• "rife4.25" / "rife4.25-tensorrt" / N/A
• "rife_elexor" / "rife_elexor-tensorrt" / N/A

Scene Change Detection

Argument	Type	Default	Description
--scenechange	flag	False	Enable scene change detection
--scenechange_sens	float	50	Sensitivity of scene change detection (0.80 - 0.90)
--scenechange_method	str	"maxxvit-directml"	Scene change detection method

Scene Change Methods

• "maxxvit-directml"
• "maxxvit-tensorrt"

Deduplication

Argument	Type	Default	Description
--dedup	flag	-	Enable deduplication
--dedup_method	str	"ssim"	Deduplication method
--dedup_sens	float	50	Sensitivity of deduplication

Dedup Methods

• "ssim"
• "mse"
• "ssim-cuda"
• "flownets"

Restoring

Argument	Type	Default	Description
--restore	flag	-	Enable denoising
--restore_method	str	"scunet"	Denoising model

Restore Methods

• "scunet"
• "nafnet"
• "dpir"
• "real-plksr"
• "anime1080fixer" / "anime1080fixer-tensorrt"
• "fastlinedarken" / "fastlinedarken-tensorrt"

Sharpening

Argument	Type	Default	Description
--sharpen	flag	-	Enable sharpening
--sharpen_sens	float	50	Sensitivity of sharpening

Segmentation

Argument	Type	Default	Description
--segment	flag	-	Enable segmentation
--segment_method	str	"anime"	Segmentation method

Segment Methods

• "anime"
• "anime-tensorrt"

Auto Cut Clips Detection

Argument	Type	Default	Description
--autoclip	flag	-	Enable auto cut clip detection
--autoclip_sens	float	40	Sensitivity of auto cut clip detection

Depth Map Generation

Argument	Type	Default	Description
--depth	flag	-	Generate depth maps
--depth_method	str	"small"	Depth map generation method
--depth_quality	str	"high"	Depth map generation quality

Depth Methods

• "small_v2" / "small_v2-tensorrt" / "small_v2-directml"
• "base_v2" / "base_v2-tensorrt" / "base_v2-directml"
• "large_v2" / "large_v2-tensorrt" / "large_v2-directml"

Depth Quality selection

• "low" / "high"
• Additional Notes: This is not reliable with TensorRT and should only be preferred if you own a low end GPU.

Resizing

Argument	Type	Default	Description
--resize	flag	-	Enable resizing
--resize_factor	float	2	Factor for resizing (>1 for upscaling, <1 for downscaling)
--resize_method	str	"bicubic"	Resizing method

Resize Methods

• "fast_bilinear", "bilinear", "bicubic", "experimental", "neighbor", "area"
• "bicublin", "gauss", "sinc", "lanczos", "spline", "spline16", "spline36", "point"

Preset Arguments

Argument	Type	Default	Description
--preset	string	-	A new and semi-baked in preset configuration system with reusability in mind.
--preset_list	flag	False	Prints all of the available presets; this comes with a forced full exit of the script after execution

Usage: --preset {name} will allow for all of the arguments declared within the initial system arguments to be saved inside a JSON file and then reused.

Note: This lacks an updating/removal system for now, so it's a one-off.

Encoding Settings

Argument	Type	Default	Description
--encode_method	str	"x264"	Encoding method
--custom_encoder	str	""	Custom FFmpeg encoding parameters

Encode Methods

Standard Encoders

• "x264": libx264, preset veryfast, CRF 15
• "x264_10bit": libx264, preset veryfast, CRF 15, high10 profile
• "x264_animation": libx264, preset veryfast, tune animation, CRF 15
• "x264_animation_10bit": libx264, preset veryfast, tune animation, CRF 15, high10 profile
• "x265": libx265, preset veryfast, CRF 15
• "x265_10bit": libx265, preset veryfast, CRF 15, main10 profile
• "h266": libvvenc, QP 24, preset 0

Hardware Encoders

• "nvenc_h264": h264_nvenc, preset p1, CQ 15
• "nvenc_h265": hevc_nvenc, preset p1, CQ 15
• "nvenc_h265_10bit": hevc_nvenc, preset p1, CQ 15, main10 profile
• "nvenc_av1": av1_nvenc, preset p1, CQ 15
• "qsv_h264": h264_qsv, preset veryfast, global quality 15
• "qsv_h265": hevc_qsv, preset veryfast, global quality 15
• "qsv_h265_10bit": hevc_qsv, preset veryfast, global quality 15, main10 profile
• "qsv_vp9": vp9_qsv, preset veryfast, global quality 15
• "h264_amf": h264_amf, quality speed, RC CQP, QP 15
• "hevc_amf": hevc_amf, quality speed, RC CQP, QP 15
• "hevc_amf_10bit": hevc_amf, quality speed, RC CQP, QP 15, main10 profile

Other Formats

• "av1": libsvtav1, preset 8, CRF 15
• "prores": prores_ks, profile 4, qscale 15
• "gif": gif, qscale 1, loop 0
• "image": png, quality 1
• "vp9": libvpx-vp9, preset veryfast, CRF 15

High Quality Presets

• "slow_x264": libx264, preset slow, CRF 18, high profile, level 4.1, tune animation, x264-params ref=4:bframes=8:b-adapt=2:direct=auto:me=umh:subme=10:merange=24:trellis=2:deblock=-1,-1:psy-rd=1.00,0.15:aq-strength=1.0:rc-lookahead=60, bf 3, g 250, keyint_min 25, sc_threshold 40, qcomp 0.6, qmin 10, qmax 51, maxrate 5000k, bufsize 10000k, movflags +faststart
• "slow_x265": libx265, preset slow, CRF 18, profile main, level 5.1, tune ssim, x265-params ref=6:bframes=8:b-adapt=2:direct=auto:me=umh:subme=7:merange=57:rd=6:psy-rd=2.0:aq-mode=3:aq-strength=0.8:rc-lookahead=60, bf 4, g 250, keyint_min 25, sc_threshold 40, qcomp 0.7, qmin 10, qmax 51, maxrate 5000k, bufsize 10000k, movflags +faststart
• "slow_av1": libsvtav1, preset 4, CRF 30, VBR, maxrate 5000k, bufsize 10000k, tile-columns 2, tile-rows 2, row-mt 1, movflags +faststart

Note: The output pixel format (-pix_fmt) can dynamically switch between yuv420p, yuv420p10le, and yuv444p10le based on --bit_depth and --encode_method.

Performance Settings

Argument	Type	Default	Description	Choices
--half	string	True	Enable FP16 for improved performance	True, False
--buffer_limit	int	50	Set the limit for decoding and encoding buffers
--static	flag	True	Force static engine generation and bypass the arbitrary set limitations of TAS
--decode_threads	int	4	The amount of threads utilized for decoding

Note: Interpolation TRT, Depth TRT, Segment TRT all already are set to static engine generation due to set limitations within the models themselves.

Usage Examples

Note: Replace .\input\test.mp4 with a full or relative path to a video file or folder. The --output argument is optional, as the script can automatically generate a file name based on the input.

When using the packaged main.exe file, replace python .\main.py with main.exe in the following examples.

1. Enable upscaling

python .\main.py --input .\input\test.mp4 --upscale

2. Enable interpolation with Rife 4.6:

python .\main.py --input .\input\test.mp4 --interpolate --interpolate_method rife4.6

3. Enable upscaling and interpolation with custom models:

python .\main.py --input .\input\test.mp4 --upscale --upscale_method superultracompact --interpolate --interpolate_method rife4.20

4. Enable deduplication with custom sensitivity and method:

python .\main.py --input .\input\test.mp4 --dedup --dedup_sens 60 --dedup_method ssim

5. Enable denoising using the NAFNet model:

python .\main.py --input .\input\test.mp4 --restore --restore_method nafnet

6. Enable interpolation with scene change detection:

python .\main.py --input .\input\test.mp4 --interpolate --scenechange

7. Generate depth maps using the "large" method and 16-bit workflow:

python .\main.py --input .\input\test.mp4 --depth --depth_method large_v2 --bit_depth 16bit

8. Download a YouTube video, upscale with Span-TensorRT, interpolate with Rife 4.17-TensorRT, and use scene change detection:

python .\main.py --input https://www.youtube.com/watch?v=dQw4w9WgXcQ --upscale --upscale_method span-tensorrt --interpolate --interpolate_method rife4.17-tensorrt --scenechange

(Replace the YouTube URL with the actual video URL you want to download)

9. Process video from second 8 to 17 with custom settings:

python .\main.py --input .\input\test.mp4 --inpoint 8 --outpoint 17 --upscale --upscale_method superultracompact-tensorrt --interpolate --interpolate_method rife4.20 --scenechange --scenechange_method maxxvit-tensorrt --custom_encoder "-c:v libsvtav1 -preset 8 -crf 30 -vf hqdn3d=1.5:1.5:6:6 -pix_fmt yuv420p"

10. Upscale, Interpolate, Scenechange, x264_animation_10bit and 16bit workflow example:

python .\main.py --input https://www.youtube.com/watch?v=YRL74JmhVgk --upscale --upscale_method shufflecugan-tensorrt --interpolate --interpolate_method rife4.22-tensorrt --scenechange --scenechange_method maxxvit-tensorrt --encode_method x264_animation_10bit --bit_depth 16bit