Recently tested the Everlyn-1 video generation tool, and it indeed produces videos in just 16 seconds—fast beyond belief. But initially, I kept hitting a question: where exactly are the training data and finished videos stored? Looking at the AWS bills, the growth rate is even more aggressive than the model iterations. Later, I found out that their entire storage system is built on a top-tier blockchain storage protocol—once I dug into it, I realized that the training sets, checkpoints, and KV caches are all migrated there. They use a coding technology called Red‑Stuff, which can push storage costs down to almost nothing. For a team like ours that produces hundreds of test clips daily, this cuts costs by more than half.

The most surprising part was how they handle fragmented files. Previously, batch storing small files was slow and prone to frame loss. Their Quilt solution can pack scattered frame sequences for quick storage and automatically restore the original structure during retrieval. It even saves preprocessing time. Coupled with integration with the Sui chain—profits are settled instantly via smart contracts, and data notarization is uploaded to the chain in real-time—making creative work surprisingly reliable.

Now I finally understand: decentralized storage isn’t just about having an extra backup; it has become the core of the AI video workflow. Costs are lowered, efficiency is increased, and the entire ecosystem can collaborate more tightly. I hear they plan to open interfaces to third parties later on, so the various plugins and tools we develop might also run directly on this infrastructure.