🎬 The Anime Scripter (TAS)

High-performance AI video enhancement toolkit for creators

Free, open-source anime video toolkit: AI upscale, RIFE interpolation and restore in one pass.

720p master → Adore 2x. More demos · Promo video

Why TAS vs Topaz / FlowFrames / chaiNNer
Anime-first + free	CUGAN / Adore / SPAN / Fallin tuned for line art, CLI free (AGPL) vs $299/yr
One pass, fast	Dedup → interpolate → upscale → restore in memory, with TensorRT / CUDA graphs
Where you edit	Only free tool with native After Effects panel + Standalone + scriptable CLI

📋 Overview

TheAnimeScripter (TAS) is a cutting-edge AI-powered video enhancement toolkit specialized for anime and general video content, bringing professional-grade AI upscaling, interpolation, and restoration to creators.

TAS ships in three editions:

Edition	What it is	For
TAS-Standalone	Windows desktop app with its own bundled backend — no Python, no command line	Most users
TAS-AdobeEdition	After Effects plugin, runs TAS without leaving AE	AE workflows
CLI	main.py / tas — every parameter, scriptable	Power users, automation

📖 Table of Contents

• 📋 Overview
• 🚀 Key Features
• 🖥️ User Interfaces
• 🛠️ Getting Started
• 📚 Available Inputs
• 📁 Available Models
• 🙏 Project Contributors
• 🌟 Star History
• 👥 Code Contributors

🚀 Key Features

🎞️ Video Enhancement Motion Interpolation: Create buttery-smooth animation with advanced frame interpolation AI Upscaling: Enhance resolution with AI-powered upscaling (2x) Motion Blur: Simulate realistic camera motion blur via interpolation and weighted frame blending Smart Deduplication: Optimize file size and interpolation by removing redundant frames	🎭 Advanced Editing Background-Foreground Segmentation: Precise automatic rotobrushing Depth Map Generation: 3D-ready depth maps for creative effects Object Detection: YOLOv9-based detection with bounding boxes, masks, and class labels Video Stabilization: SuperPoint feature-matching stabilizer for shaky footage
🔧 Workflow Optimization TAS-Standalone Desktop App: One-installer Windows GUI with presets, live progress, and themes After Effects Integration: Seamless plugin for AE workflow Model Chaining: Combine multiple effects in a single processing pass In-Memory Processing: Efficient frame handling without redundant disk operations	🧠 AI Flexibility Multi-Backend Support: CUDA, TensorRT, DirectML, and OpenVINO acceleration Custom Model Support: Import your own trained models Restoration Options: Denoise, dejpeg, sharpen, and line enhancement

🖥️ User Interfaces

Adobe After Effects Integration

Currently being enhanced with new features and optimizations

Adobe plugin provides seamless integration directly within your AE workflow, enabling AI-powered video enhancement without leaving your editing environment.

Windows Desktop App (TAS-Standalone)

The native Windows desktop app. It ships with its own backend, so nothing else has to be installed — no Python, no CLI, no After Effects. Build a processing chain by toggling steps, watch the live preview and logs while it runs, and save the setup as a preset.

🛠️ Getting Started

TAS-Standalone

The easiest way to use TAS. A single Windows installer that includes the full TAS backend — you do not need Python, the CLI, or After Effects.

System Requirements:

• Windows 10 / 11 (64-bit)
• Compatible GPU recommended:
  • Modern NVIDIA (RTX 20/30/40, GTX 16 series): Full CUDA/TensorRT support
  • Older NVIDIA (GTX 1000 series/Pascal): DirectML backend
  • Intel dGPU / iGPU: OpenVINO backend
  • AMD/Intel: DirectML backend

Installation Steps:

1. Download TAS-<version>-Standalone-Windows-Setup.exe from the releases page
2. Run the installer
3. Launch TAS-Standalone, pick an input video, choose your options, and hit run

Models are downloaded on first use of each option, so the initial run of a given model takes longer.

Adobe After Effects Plugin

System Requirements:

• After Effects 2022 or higher
• Compatible GPU recommended:
  • Modern NVIDIA (RTX 20/30/40, GTX 16 series): Full CUDA/TensorRT support
  • Older NVIDIA (GTX 1000 series/Pascal): DirectML backend
  • Intel dGPU / iGPU: OpenVINO backend
  • AMD/Intel: DirectML backend

Installation Steps:

1. Download the TAS-AdobeEdition from the releases page
2. Extract the .zip file to a location of your choice
3. Follow the installation tutorial to add TAS to After Effects

Need help? Watch the video tutorial

Command Line Interface

Get the most powerful and flexible version of TAS with the command-line interface:

• ⬇️ Stable Release — Recommended for production work
• ⬇️ Nightly Builds — Latest features (may contain bugs)

For local development, install the base requirements and the runtime profile you want with pip:

python -m pip install -r requirements.txt -r extra-requirements-windows-lite.txt
python main.py -h

On Apple Silicon macOS, use the MPS profile:

python -m pip install -r requirements.txt -r extra-requirements-macos.txt
python main.py -h

macOS needs a Homebrew FFmpeg. TAS does not ship FFmpeg — it is GPL, and bundling it would make every TAS release a redistributor — so on first run TAS installs it on your machine with brew install ffmpeg if it is not already present. Homebrew itself must be installed; TAS will not install it for you.

Swap extra-requirements-windows-lite.txt for the matching Windows/Linux CUDA/lite profile or extra-requirements-macos.txt for Apple Silicon.

Windows CLI Installer

Use this one-liner:

iwr -useb https://tas.nevermindnilas.dev/install.ps1 | iex

It installs into a TheAnimeScripter folder inside the directory you launch it from and prompts whether that folder should be added to your user PATH.

After installation, you can use tas --help or theanimescripter --help.

📚 Available Inputs

All available parameters for interacting with the CLI or directly with main.py can be found in the Parameters guide.

📁 Available Models

🆙 Upscaling Models

Model	CUDA	TensorRT	DirectML	OpenVINO
Cyte V1 (2×)	✅	✅	✅	✅
ShuffleCugan	✅	✅	✅	✅
Fallin Soft	✅	✅	✅	✅
Fallin Strong	✅	✅	✅	✅
Span	✅	✅	✅	✅
OpenProteus	✅	✅	✅	✅
AniScale 2	✅	✅	✅	✅
RTMOSR	❌	✅	✅	✅
Saryn ( RTMOSR based )	✅	✅	✅	✅
Gauss ( DIS based )	✅	✅	✅	✅
Custom (Spandrel)	✅	❌	❌	❌
NVIDIA Maxine VSR	✅	✅	❌	❌

⏱️ Interpolation Models (RIFE)

Version	CUDA	TensorRT	DirectML	OpenVINO
4.6	✅	✅	✅	✅
4.15	✅	✅	✅	✅
4.15-lite	✅	✅	✅	✅
4.16-lite	✅	❌	❌	❌
4.17	✅	✅	✅	✅
4.18	✅	✅	✅	✅
4.20	✅	✅	✅	✅
4.21	✅	✅	✅	✅
4.22	✅	✅	✅	✅
4.22-lite	✅	✅	✅	✅
4.25	✅	✅	✅	✅
4.25-heavy	✅	✅	✅	✅
Rife_Elexor (mod 4.7)	✅	✅	✅	✅

🔧 Restoration Models

Model	CUDA	TensorRT	DirectML	OpenVINO
SCUNet (Denoise)	✅	✅	❌	❌
NAFNet (Denoise)	✅	❌	❌	❌
DPIR (Denoise)	✅	❌	❌	❌
DeJpeg ( Real-Plksr )	✅	❌	❌	❌
Anime1080fixer	✅	✅	✅	✅
FastLineDarken	✅	✅	❌	❌
GaterV3	✅	❌	✅	✅
DeH264 ( Real-Plksr )	✅	✅	✅	✅
deepDeband-f (Debanding)	✅	❌	❌	❌
NVIDIA Maxine (Denoise/Deblur)	✅	✅	❌	❌

🌊 Depth Map Models

Model	CUDA	TensorRT	DirectML	OpenVINO
"Faster" Depth-Anything v2 Models
Small v2	✅	✅	✅	✅
Original Implementation Models
OG Small v2	✅	✅	❌	✅
Limbo Models (anime)
Limbo v1 (--depth_method limbo)	✅	✅	❌	✅
Limbo v2 (--depth_method limbo_v2)	✅	✅	❌	✅

DA3 depth-video streaming is available on CUDA as video_small_v3 and video_base_v3, reusing the Apache-2.0 Small/Base checkpoints. For example:

python main.py --input input.mp4 --output depth.mp4 --depth_method video_small_v3 --depth_window 8

Limbo V1 and V2 also support CUDA streaming as video_limbo and video_limbo_v2. They reuse the existing Limbo checkpoints and fixed input resolutions; --depth_quality is ignored for these modes.

Frames attend across overlapping chunks; --depth_window sets the chunk size (4/8/16/32, default 32), with 50% overlap. Smaller windows reduce memory and lookahead latency. This is a depth-only adaptation of DA3-Streaming, without camera reconstruction or loop closure. See details and licenses.

🎯 Object Detection Models (YOLOv9-MIT)

Model	CUDA	TensorRT	DirectML	OpenVINO
YOLOv9 Small	✅	✅	✅	✅
YOLOv9 Medium	✅	✅	✅	✅
YOLOv9 Large	✅	✅	✅	✅

🎥 Video Stabilization

Method	CUDA
SuperPoint feature matching	✅

🙏 Project Contributors

🧠 Model & Algorithm Contributors

Contributor	Contribution	Repository
SUDO	ShuffleCugan & ONNX models	VSGAN-tensorrt-docker
renarchi	Adore, Fallin Soft & Strong models	Fallin-Upscale
HZWER	RIFE interpolation framework	Practical-RIFE
SkyTNT	Anime segmentation models	anime-segmentation
DepthAnything	Depth map generation	Depth-Anything-V2
Sirosky	Open-Proteus & AniScale 2	Upscale-Hub
Elexor	Custom RIFE modifications	Modded Rife Experiment(s)
Zarxrax	Anime1080Fixer restoration	GitHub
umzi	RTMOSR & GaterV3 Models	GitHub
Phhofm	DeJpeg & DeH264 restoration	Phhofm/models
Kim2091	DIS Architecture (Gauss)	DIS
Raymond Zhou et al.	deepDeband (ICIP 2022) debanding	deepDeband

🛠️ Framework & Tool Contributors

Contributor	Contribution	Repository
FFmpeg Group	Video processing framework	FFmpeg
YT-DLP Team	Media download capabilities	yt-dlp
Breakthrough	Scene detection algorithms	PySceneDetect
Chainner-org	Neural network architecture (vendored fork)	spandrel
TNTWise	RIFE ONNX optimizations & NCNN model fork	rife-ncnn-vulkan
Media2x	RIFE NCNN Python (SWIG) wrapper	rife-ncnn-vulkan-python
nihui	RIFE NCNN Vulkan implementation	rife-ncnn-vulkan
Hyperbrew	Adobe integration framework	bolt-cep
NVIDIA	Maxine Video Effects SDK (RTX VSR) via nvidia-vfx	NVIDIA Maxine

🌟 TAS Collaborators

Contributor	Contribution
Trentonom0r3	TAS Adobe Edition
Adegerard	Project architecture & optimization suggestions

📧 Missing contributor? Please contact me at nilascontact@gmail.com or open a Github Issue!

🌟 Star History

👥 Code Contributors

A huge thank you to everyone who has contributed code to TheAnimeScripter.