WARNING: this is not a lora its a ref mod!
its realy easy to do ref mod
no training or anything like prompting
im going to say it
no single model can do anime as well as minimax h3 even seedance2.5
i found out minimax h3 is much better in ref mods than loras
i mean very better
super accurate to orginal
download these 2 nodes:
https://github.com/FranckyB/ComfyUI-H3RefModPicker
https://github.com/Luisacaotica/ComfyUI-MiniMaxH3Mod
---
load them like how you load loras
---
version 2.0:
the right eye was fixed (it was invisible)
Description
FAQ
Comments (12)
do you have any guide or tutorial on how can we do that as I remember the audio in the latent is not possible.
Yes you can
Minimax h3 is a powerfull model you can add a audio in latent with "create refmod from folder" node in that node pack that i put in the discription.
This rem refmod has about 10 image refrences and 2 audios
I dont want to sound mean or discouraging but refmods are not loras and you should not put tag them as such. They're more closer to workflows than loras. Anything refmod does, a simple ref workflow (but actually refmod is worse as you lose reference model syntax flexibility) can do. Image, video, or audio concept is the same.
True, it should be a refmod tag here.
there is no tag for it
its safe tensor so it also wont be counted as a workflow
i hope for a refmod tag
i see a other tag
so here we are
thank you for your comment
also i think loras is much more worse in flexibility than refmods
refmod keeps the model weights and doesnt destory them
in low values its good but still you lose the closeness
thats why for characters i think ref mod is better
@to_MacDonalds_and_beyond The only advantage you save is the encoding time for the input data. Adding to the wrong category is not helping anyone. It's going to just be a mess when people add tons of these on the lora page. And loras will always be more flexible, the issue right now is training is difficult for H3 as there is no base model to train from. I hope they can add a refmod category, that would be very helpful. Since there is a use for them though.
@tazmannner379 That’s their advantage:
They don’t release the base version.
The reason is obvious:
Better fine-tuned models would emerge, and they would lose control over the core model.
By the way, I changed the category from LoRA to "Other."
dont be a jerk guys
its simplely a refmod
this is just a showcase to show how powerfull it is😊
@to_MacDonalds_and_beyond Most of the time, people just complain about what others share. If you don't like it and you have nothing constructive to say, well, just move on guys.
Personnally, I'll take it with pleasure, thank you for making Refmods and saving my time 💜
@K0DEX i just made this so people can see this new metod and those with low pc specs can make their own characters without some training
@to_MacDonalds_and_beyond and I thank you for doing this, otherwise I probably wouldn't have taken a look at RefMods, or much later.