LatentBox cuts AI image storage by 78.7% with latent-first approach

The explosive growth of AI-generated images has created a sustainability crisis for storage infrastructure. Platforms like Midjourney and Adobe Firefly host billions of generative images, but conventional object stores persist them as full-resolution pixel blobs, consuming massive storage and bandwidth. Unlike natural photos, AI-generated images can be deterministically reconstructed from compact, model-native latent tensors, making persistent pixel storage fundamentally redundant. In response, researchers from multiple universities developed LatentBox, a latent-first storage system that treats compressed latents as durable storage objects and uses on-demand GPU reconstruction on the read path to trade inexpensive compute for large persistent storage savings.

The design is guided by the first large-scale analysis of AI-generated image access, based on a 35-month, 2-billion-request production trace from a major generative-content platform. LatentBox keeps frequently accessed images in decoded pixel format for fast hits, stores less-active objects as compressed latents to expand effective cache capacity, and continuously adjusts the split between the image and latent caches to optimize user-perceived access latency. In prototype evaluation using the production trace, LatentBox reduces persistent storage by 78.7% with competitive or even lower mean and tail latency compared to a pure image-based storage system.