Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Defining an Efficient and Expandable File Format for AI-Generated Contents

Yixin Gao, Runsen Feng, Xin Li, Weiping Li, Zhibo Chen

TL;DR

AI-generated content creates vast volumes of high-fidelity images, straining storage and transmission. The authors propose AIGIF, a lossless image format that compresses the generation syntax—platform, model, and data configurations—rather than pixel data, achieving ultra-low bitrate performance. Through a systematic study of platform, model, and data factors, they design a composable bitstream with an expandable exp code to accommodate future models. Empirically, AIGIF reaches up to $1/10{,}000$ compression with high fidelity and provides a robust direct-pixel fallback, enabling scalable, cross-platform reproducibility for AIGC pipelines.

Abstract

Recently, AI-generated content (AIGC) has gained significant traction due to its powerful creation capability. However, the storage and transmission of large amounts of high-quality AIGC images inevitably pose new challenges for recent file formats. To overcome this, we define a new file format for AIGC images, named AIGIF, enabling ultra-low bitrate coding of AIGC images. Unlike compressing AIGC images intuitively with pixel-wise space as existing file formats, AIGIF instead compresses the generation syntax. This raises a crucial question: Which generation syntax elements, e.g., text prompt, device configuration, etc, are necessary for compression/transmission? To answer this question, we systematically investigate the effects of three essential factors: platform, generative model, and data configuration. We experimentally find that a well-designed composable bitstream structure incorporating the above three factors can achieve an impressive compression ratio of even up to 1/10,000 while still ensuring high fidelity. We also introduce an expandable syntax in AIGIF to support the extension of the most advanced generation models to be developed in the future.

Towards Defining an Efficient and Expandable File Format for AI-Generated Contents

TL;DR

AI-generated content creates vast volumes of high-fidelity images, straining storage and transmission. The authors propose AIGIF, a lossless image format that compresses the generation syntax—platform, model, and data configurations—rather than pixel data, achieving ultra-low bitrate performance. Through a systematic study of platform, model, and data factors, they design a composable bitstream with an expandable exp code to accommodate future models. Empirically, AIGIF reaches up to compression with high fidelity and provides a robust direct-pixel fallback, enabling scalable, cross-platform reproducibility for AIGC pipelines.

Abstract

Recently, AI-generated content (AIGC) has gained significant traction due to its powerful creation capability. However, the storage and transmission of large amounts of high-quality AIGC images inevitably pose new challenges for recent file formats. To overcome this, we define a new file format for AIGC images, named AIGIF, enabling ultra-low bitrate coding of AIGC images. Unlike compressing AIGC images intuitively with pixel-wise space as existing file formats, AIGIF instead compresses the generation syntax. This raises a crucial question: Which generation syntax elements, e.g., text prompt, device configuration, etc, are necessary for compression/transmission? To answer this question, we systematically investigate the effects of three essential factors: platform, generative model, and data configuration. We experimentally find that a well-designed composable bitstream structure incorporating the above three factors can achieve an impressive compression ratio of even up to 1/10,000 while still ensuring high fidelity. We also introduce an expandable syntax in AIGIF to support the extension of the most advanced generation models to be developed in the future.

Paper Structure

This paper contains 13 sections, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminary
  4. AIGIF
  5. Observations
  6. Composable Bitstream Structure
  7. Expandable Syntax
  8. Overall Pipeline
  9. Experiments
  10. Experimental Setup
  11. Results
  12. Limitations
  13. Conclusion

Figures (3)

  • Figure 1: High-level comparison of image saving and image recreation process between (a) common image formats like PNG png and (b) our proposed image format. Rather than directly saving compressed image pixels into a file, we save the compact generation information into a file as the representation of AI-generated images. Dash line is optional.
  • Figure 2: Overview of the generation information for reproducing the image generation process.
  • Figure 3: Comparison of image generation across different models with identical hyperparameter settings.