# PyTorch moves all media I/O to TorchCodec

> The new TorchCodec library now handles decoding and encoding for images, video, and audio, while TorchVision and TorchAudio focus on transforms.

Oossa · 2026-10-05 · https://oossa.com/en/pytorch-moves-all-media-i-o-to-torchcodec

PyTorch has shifted every media decoding and encoding function into a single library called TorchCodec. The change means that TorchVision and TorchAudio no longer provide I/O APIs; they now only offer transforms for tensors. TorchCodec works on both CPU and GPU and supports many formats via FFmpeg and other codecs. All three libraries are now ABI‑stable, so they can be updated without rebuilding PyTorch.

## The facts

- TorchCodec consolidates image, video and audio I/O that was previously split across TorchVision and TorchAudio.
- The three libraries are now ABI stable, decoupling their release cycle from PyTorch.

## Why it matters

Developers can get faster, simpler media handling by using TorchCodec and rely on stable libraries for transforming data.

## Sources & references

1. [Evolution of the PyTorch Media Processing Landscape](https://pytorch.org/blog/evolution-of-the-pytorch-media-processing-landscape/) – PyTorch

Last updated: 2026-10-05