CogView2: Faster and Better Text-to-Image Generation via Hierarchical Transformers

TL;DR

CogView2 introduces a hierarchical transformer approach with local parallel auto-regressive generation, enabling faster, high-resolution text-to-image synthesis with interactive editing capabilities, outperforming some existing models.

Contribution

It proposes a novel hierarchical transformer architecture and training strategy that significantly improves speed and quality in text-to-image generation.

Findings

Competitive image generation quality compared to DALL-E-2

Supports interactive text-guided image editing

Achieves faster high-resolution image synthesis

Abstract

The development of the transformer-based text-to-image models are impeded by its slow generation and complexity for high-resolution images. In this work, we put forward a solution based on hierarchical transformers and local parallel auto-regressive generation. We pretrain a 6B-parameter transformer with a simple and flexible self-supervised task, Cross-modal general language model (CogLM), and finetune it for fast super-resolution. The new text-to-image system, CogView2, shows very competitive generation compared to concurrent state-of-the-art DALL-E-2, and naturally supports interactive text-guided editing on images.