Face and speech dynamics for Minimax H3
Full face muscle dynamics with clear English speech
Trained on T2V but supports also I2V
720x1280 resolution with 15-30 steps
Weight 0.4-0.8 is recommended
Pro Tip: to prevent voice drift, use fixed seed and start your text with the same word (hey guys,)
Prompt: he is speaking into a mic, he maintains eye contact with the viewer saying: 'text here'
Description
First initial release for Minimax H3.
FAQ
Comments (19)
This is great looking! will try it soon! also no rush or anything, but any progress on the twerking lora for minimax h3?
How well does it work with other languages?
I have trained v1 only with English because most will know it, to add multiple languages is hard because i would need to understand the language to know it works.
@AdaptiveVision It worked perfectly in brazilian portuguese for me. even at 0.8 strength. Amazing lora btw, thanks for creating it :D
@Wolphtizando That is good, no problem!
Just did a quick test with Swedish and that worked very well. 👍
Finally no more extreme mouth movements when talking.
works great in french !
Works perfect in spanish.
We need those kind of lora ! \o/
works great in french but turns all women blonde and looking very close to the one in your exemple.
i tried lower lora level, but it's not as effective ( the speech)
@tta175 ça vient pas du LoRA je penses, je le test aussi en français et j'ai pas ce problème.
That is not the LoRA issue, as you can see it works in multiple ways.
@AdaptiveVision Works great by the way, thanks for your work !
@K0DEX strange i'll make some more tests, but it was the only lora i used
The battle to fight Minimax H3 plastic skin carries on. Let's all fight this together.
Questions :
- Are you using a finetune?
- Are you using Turbo LoRAs?
I know exactly what you mean. I read that ref2 gives you more plastic look, thats why i avoid it, also not using a hybrid just fl2 pruned or unpruned does not matter in this case. Since that this problem is kinda under control for me.