Generative AI video is currently quite popular, but everyone is stuck on a common bottleneck—high storage costs, slow data retrieval, and user privacy protection remain major issues.

Recently, I came across an interesting case. The generative AI video platform Everlyn has turned its attention to decentralized storage, choosing Walrus as the core data layer, directly breaking through these industry dilemmas. This deep integration of AI and distributed storage is actually quite noteworthy.

What is Everlyn’s core competitive advantage? Their Everlyn-1 model can convert static images into high-quality videos in 16 seconds, a speed that far surpasses platforms like Midjourney. The technical support behind this is inseparable from Walrus’s backing.

This collaboration is quite substantial. Everlyn has already migrated over 5,000 user videos, ranging from 480p to 720p resolution, to Walrus. They also plan to transfer all key data such as training datasets, model checkpoints, and KV caches stored on AWS and Azure, totaling over 50GB. For an AI video generation platform, this is a major move.

Why do this? Cost. The storage cost of training data directly determines service pricing. Walrus’s Red-Stuff 2D erasure coding technology can reduce storage costs to a new industry low, allowing Everlyn to maintain high-speed generation capabilities while lowering the barrier to entry for creators.

There’s also a detail—Quilt’s batch storage solution. Video generation produces a large number of fragmented small files, and this solution can handle them perfectly, enabling efficient batch processing and fast access, ensuring real-time model optimization. It looks like a comprehensive solution.