# Integration of Jev into Minimax H3 Video Generation — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/39254aa1a9d23b35

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-21 03:35 ingested 2026-09-22 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wm1mgy/integration_of_jev_into_minimax_h3_video/)

## retained text



```
Apparently, someone genius came up with this brilliant idea of using Jev for sparsification of attention processing. The result was speed up of generation time over 40 percent, 06:07 to 03:34. Here is the github page of the author:

https://github.com/sepiablue-ai/ComfyUI-MiniMax-H3-W4A4-VSA/tree/exp/jev-adaptive-vsa

Since there are many open-source alternatives of jev at the moment, I guess we will soon have further optimizations made soon.

Has anyone tried the repo?
```



document fetched 2026-09-22 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1wm1mgy/integration_of_jev_into_minimax_h3_video
