Lora from HunyuanVideo is currently the open-source model that can make characters 99.99% similar to those in the show. I want to say it can instantly boost your training level by at least one grade, like from A to A+, or even directly to A++. I'm really curious why the previous Hunyuan image models couldn't achieve this.
Training Lora for HyV is easy, but the testing phase is really painful. We need much more powerful GPUs!
Description
FAQ
Comments (13)
Why is this tagged real person?
You can try again
how i can use this lora art ?normal
check this out: https://civitai.com/articles/11764
Excellent lora! What guidance and flow works best?
check this out: https://civitai.com/articles/11764
What was the point in making this when a better model with all of the Disney princesses has already been released?
Which one do you mean is the better version? Do you have a link? I'd like to know.
@tylerburden100 ok
this version is the best so far. Very glad you made it :)
May I ask what prompt words you are using? I used the trigger word like "Elsa's winter dress", but it didn't generate the specified clothes, it came up with random one. Is there something wrong with my usage?
My advice is you could start by setting the frame length to 1. That way, you can test your prompt really quickly. Also, you could try using a screenshot of Elsa's winter dress and use an LLM with image recognition to get a prompt description. Or you could just amp up the weight of the 'Elsa's winter dress' prompt, or the Lora's weight.
The reason this is happening is because I wanted to keep the Lora flexible. When I was training it, my goal was for the Lora to make Elsa look almost exactly like she does in the movie, and that was good enough for me. I didn't train it to the point where the Elsa appearance would always force its way in, no matter what. I think that would be overtraining. I guess that's where my philosophy differs from other trainers.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.