malcolmrey

/

minimaxh3

Like 63

33 GBs of space apparently.

Are ref loras trained on audio?

I still dont understand dofference between them ?

and how to use reflora ??

Can i use both of the same person ? or no point ?

any workflow to use reflora ?

https://huggingface.co/datasets/malcolmrey/various/blob/main/h3-center/docs/MINIMAX_H3_REFLORA_GUIDE.md

Answered here.

RefMod is just a bunch of reference images (and or audio/video but usually images work best) stored in safetensors so you don't have to do it manually and also saves times on conversion.

I don't have to explain what a Lora is 😃

But RefLora is Lora + RefMod combined together. If you load RefLora with regular lora loader, you will get the lora part only. But if you load it with RefLora loader, you will get the Lora and the RefMod together - and it boosts the likeness 😃

You can always load Lora and RefMod(s) separately, but when I train a Lora, adding 1 mb to it to get the whole package is a bargain and it just makes it so much easier to use, and still people can only use the Lora part if they wish :)