MiniMax 的音乐模型开源了，配套的 T8 也一起放出来，comfyui 那个整合包顺势更到了 V18。有人已经跑起来了吗。
问这个是因为音乐这块之前一直卡在没几个能自己部署的选择上，基本只能用在线服务，想批量出、想抠细节都被卡着。视频这一年折腾下来本地跑通早就不稀奇了，音乐一直是空着的。现在多了一个能拿到手里的模型，至少剪片子配段背景音这种活有了别的解法。
整合包这种形式我挺看重，不会配环境的人能不能一键跑起来，往往比模型本身强多少更决定有多少人真的用得上。所以真正想问的是 V18 一键起得来吗，有跑通的说一声。
The pack updates to V18 fast, I hadn’t even finished setting up the last version 
Still not clear what T8 actually does — is it an inference-side tool or something for training
Open source doesn’t mean it’ll run — music models usually aren’t light on VRAM either, wondering if an 8GB card even has a shot
If the cost of scoring music can come down, the batch-account crowd is gonna be quietly thrilled
Waiting for someone to do a comparison on Chinese lyrics — a lot of music models stumble the moment it’s Mandarin
bookmarked