Skip to content

Open-MAGVIT2

Open autoregressive image generation models

Visit official site

What open-magvit2 is

The project is an open-source series of autoregressive image generation models released by Tencent ARC Lab. It reproduces a tokenizer and adds asymmetric tokenization plus next-sub-token prediction.

Main features

  • Model scales from 300M up to 1.5B
  • Sub-vocabularies split from a large vocabulary
  • Training scripts and model configs provided

Pros & cons

✓ What we like

  • All models and code are open source
  • Training and testing on different hardware is supported

! What to watch out for

  • Setup follows a requirements file and CUDA environment
  • Next-sub-token prediction is part of the generation design

FAQ

What reconstruction result is listed?

1.17 rFID on the ImageNet 256x256 dataset.

What size range is offered?

Models from 300M to 1.5B parameters.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review