Highly Compressed Tokenizer Can Generate Without Training

TL;DR

This paper introduces a highly compressed 1D image tokenizer that enables image editing and generation through heuristic token manipulation and gradient-based optimization, without any training of generative models.

Contribution

The work presents a novel 1D image tokenizer that allows for effective image editing and generation solely via token manipulation and optimization, bypassing traditional training.

Findings

Enables fine-grained image editing through token replacement.

Supports diverse, realistic image generation without training.

Demonstrates effective inpainting and text-guided editing.

Abstract

Commonly used image tokenizers produce a 2D grid of spatially arranged tokens. In contrast, so-called 1D image tokenizers represent images as highly compressed one-dimensional sequences of as few as 32 discrete tokens. We find that the high degree of compression achieved by a 1D tokenizer with vector quantization enables image editing and generative capabilities through heuristic manipulation of tokens, demonstrating that even very crude manipulations -- such as copying and replacing tokens between latent representations of images -- enable fine-grained image editing by transferring appearance and semantic attributes. Motivated by the expressivity of the 1D tokenizer's latent space, we construct an image generation pipeline leveraging gradient-based test-time optimization of tokens with plug-and-play loss functions such as reconstruction or CLIP similarity. Our approach is demonstrated for inpainting and text-guided image editing use cases, and can generate diverse and realistic samples without requiring training of any generative model.