ONYX AI Matte v3.3.0 视频抠像插件，AI驱动高效遮罩提取

ONYX AI Matte 是GPU加速的AI视频抠像插件，适用于达芬奇、Fusion等软件。支持文本、点、框等多模式提示，精准处理毛发、运动模糊边缘，多目标追踪，大幅提升VFX制作效率。

原始标题：ONYX AI Matte v3.3.0 x64

ONYX Ai Matte 是一款由 Evgeniy Shatskiy 开发的 GPU 加速 AI 视频抠像与遮罩提取（Matting）插件。它专为视觉特效（VFX）艺术家打造，旨在省去繁琐的手工转描（Roto）工作，能高精度处理头发、毛发、运动模糊和半透明边缘。

主要功能与特点

• 多模式提示：支持点（Point）、框（Box）、文本（Text）和遮罩（Mask）提示来引导 AI 生成遮罩。

• 多目标追踪：多数模式下可同时追踪多达 32 个对象，文本模式下最高可达 64 个。

• 复杂边缘处理：在快速运动、毛发或半透明材质的边缘能保持出色的稳定性和抠像质量。

File size: 3.69 GB

Professional AI Matte Extraction Plugin for OFX
ONYX Ai Matte is a hardware-accelerated matting plugin for DaVinci Resolve, Fusion Studio, Nuke, Flame and Natron — on Windows, Linux and macOS — built for VFX artists who need production-quality mattes without hours of manual roto work. Powered by Meta SAM3 with the VitMatte and MEMatte refiners, it runs one unified pipeline: text, point, box and mask prompts work together in a single scene, and objects track across frames automatically.

KEY FEATURES
Unified Pipeline — Box, Point, Text and Mask prompts in one scene, up to 16 objects
Text Prompts — Type any object class (or several, comma-separated) to detect all instances automatically
Point Selection — Click to segment a specific object; positive/negative points for refinement
Bounding Box — Draw a box for quick rectangular selection
Mask Prompt Input — Track a mask you already have: a one-frame roto becomes a fully tracked matte (works for sub-objects too — a sleeve, a car door); move the mask mid-track and a correction key is set right on that frame
Object Tracking — Masks propagate across frames automatically; four tracking modes for any shot
Object Manager — On-canvas panel to solo, keyframe visibility, refine and delete objects
Trimap + HQ Refiner — Automatic trimap from the masks; VitMatte/MEMatte refine the edges
Refiner Direct Input — Connect any matte (your roto, or a mask tracked in Mocha or another system) straight to the refiner for cinema-quality alpha edges
Dual Refiners — VitMatte (default) and MEMatte, switchable instantly — pick the best for your footage
TensorRT Acceleration — 3-4x faster refiner inference on NVIDIA RTX GPUs
Temporal Smoothing & Edge Stabilization — Reduce flickering across frames for clean motion
Edge Controls — Feather, shrink/grow, offset, levels, binary threshold for fine-tuning
Overlay Preview — Real-time mask preview with customizable colors
Bake to Disk — Cache masks to disk; reopen the project and play back with no engine and no VRAM

PERFECT FOR
Multi-character scenes: find all people, solo specific actors
Complex hair and fur extraction with cinema-quality edges
Fast commercial turnaround with text-first discovery
4K+ compositing in Nuke, DaVinci Resolve, Fusion Studio, Flame and Natron
Edge-refining external mattes (roto, planar-tracked) via the Refiner Direct workflow

System requirements: / TECHNICAL SPECS
OS: Windows 10/11 64-bit
AI Models: Meta SAM3 + VitMatte + MEMatte (ONNX, FP16)
Acceleration: NVIDIA CUDA 12.6 + TensorRT 10.x on Windows / Linux — Apple Neural Engine (CoreML) on macOS
Driver: NVIDIA R560 or newer (CUDA 12.6) — Windows / Linux only
VRAM: 6GB minimum / 8GB+ recommended — no separate GPU needed on macOS
RAM: 16GB minimum / 32GB for 4K
GPU: NVIDIA RTX 20-series or newer (RTX 2060+, SM 7.5+) — Windows / Linux
OS: Linux (glibc 2.28+: Ubuntu 20.04+, Debian 10+, Rocky/RHEL/AlmaLinux 8+) · macOS 15 Sequoia or newer on Apple Silicon — M1, M2, M3, M4, M5 or newer
Hosts: DaVinci Resolve, Fusion Studio, Nuke, Flame, Natron — any OFX-compatible app (Nuke 13+ on Windows and Linux, Nuke 15+ on macOS)
Not supported: GTX 1000-series, AMD Radeon, Intel Arc — and Intel-based Macs