AI Assisted Matching

AI-assisted matching — optional (v1.3.0)

You do not need AI matching. In v1.3.0, when both videos are present AutoSync matches by picture, and that engine gives the same result with Standard or AI-assisted selected (measured on all 64 test runs). AI-assisted matching only helps the older audio-matching engine, which is used when a video is missing.

AI-assisted is a matching method, not an output mode: you still choose Full target audio, Main episode only or Keep opening & ending separately.

Install the AI models

1. Open Settings (top right).
2. Under Engine & AI models, click Install AI Sync models. This downloads optional PyTorch/Transformers packages and can be large.
3. Keep internet access: the model weights download on the first AI run and are cached afterwards.

Install subtitle AI is a different button, for subtitle workflows only.

If your graphics card cannot run the installed GPU build (for example some RTX 50-series cards with older CUDA builds), AutoSync detects it and uses the CPU instead of failing.

Use it

In the Audio tab, under Matching → Method, choose AI-assisted (experimental), then start as usual.

What it does

• An Audio Spectrogram Transformer (AST) proposes matching points between the two soundtracks.
• DINOv2-small scene matching can help where conventional scene matching is weak.
• AutoSync checks every proposed point against the original sound and applies the usual safety rules.

← Wiki home · Getting started