Cover image

NUWA Infinity

A multimodal generative foundation model that is designed to generate high-quality images and videos from given text, image or video input.

NUWA-Infinity, a generative model for infinite visual synthesis, which is defined as the task of generating arbitrarily sized high-resolution images or long-duration videos.  An autoregressive over autoregressive generation mechanism is proposed to deal with this variable-size generation task, where a global patch-level autoregressive model considers the dependencies between patches, and a local token-level autoregressive model considers dependencies between visual tokens within each patch. A Nearby Context Pool (NCP) is introduced to cache-related patches already generated as the context for the current patch being generated, which can significantly save computation costs without sacrificing patch-level dependency modeling. An Arbitrary Direction Controller (ADC) is used to decide suitable generation orders for different visual synthesis tasks and learn order-aware positional embeddings. Compared to NUWA, which also covers images and videos, NUWA-Infinity has superior visual synthesis capabilities in terms of resolution and variable-size generation.

Please visit our homepage to read more > (opens in new tab)