Language-based Image Colorization: A Benchmark and Beyond

Wangxuan Institute of Computer Technology, Peking University    

Figure 1. Taxonomy of automatic and language-based image colorization methods.


Abstract

Image colorization aims to bring colors back to grayscale images. Automatic image colorization methods, which requires no additional guidance, struggle to generate high-quality images due to color ambiguity, and provides limited user controllability. Thanks to the emergency of cross-modality datasets and models, language-based colorization methods are proposed to fully utilize the efficiency and flexibly of text descriptions to guide colorization. In view of the lack of a comprehensive review of language-based colorization literature, we conduct a thorough analysis and benchmarking. We first briefly summarize existing automatic colorization methods. Then, we focus on language-based methods and point out their core challenge on cross-modal alignment. We further divide these methods into two catagories: one attempts to train a cross-modality network from scratch, while the other utilizes the pre-trained cross-modality model to establish the textual-visual correspondence. Based on the analyzed limitations of existing language-based methods, we propose a simple yet effective method based on distilled diffusion model. Extensive experiments show that our simple baseline can produces better results than previous complex methods with 14 times speed up. To the best of our knowledge, this is the first comprehensive review and benchmark on language-based image colorization field, providing meaningful insights for the community.

Summary of main features

We summarize main feautures of language-based methods in a brief sentence.

Diffusion-based methods analysis

We summarize four representative condition insertion paradigms of existing Diffusion-based methods. These four methods can constraint the generation process by grayscale images with different injection intensities and feature granularity.

Model architecture of our Color-Turbo.

Although the pipeline is simple, it is effctive to generate high-quality results with fast inference speed!

Visual Results

Some results of existing language-based methods and our Color-Turbo. You can zoom in for better visualization!

Quantitative Benchmark Results

Some results of existing language-based methods and our Color-Turbo. You can zoom in for better visualization!

BibTeX

Please consider to cite LIC-Benchmark and Color-Turbo if it helps your research.
@article{li2025colorturbo,
  title={Language-based Image Colorization: A Benchmark and Beyond},
  author={Li, Yifan and Yang, Shuai and Liu, Jiaying},
  booktitle={arXiv: 2503.14974},
  year={2025}
}