Reward lora gives same face.
Dear vrgamedevgirl84. I'm not sure if you're aware of this but the reward lora gives "same face". Meaning the persons faces has a tendency to stay the same with it enabled or merged into your model. The face just looks overly smoothed and a little bit fake. Just thought you should know this. Maybe remerging with a lower lora weight or tell people to subtract themselves by the desired amount in their workflows could be an option.
Otherwise thank you so very much for contributing to the ai community. :)
Thanks for your feedback! I did set the strength lower when merging—any lower and it wouldn't have had any effect. That LoRA is one of the key ones that really helps make FusionX what it is. What settings are you using in your WF? Personally, I don't have an issue with the "same face" thing, but that could be down to my prompts.
There's also a new FusionX model on the way—it’s a FusionX LoRA, so you’ll be able to adjust its strength, which might help with this. If you join the Discord, you can see some examples and maybe send me a sample image to test. The link’s in the model card description.
I made Space REPO(souce code include): https://huggingface.co/spaces/Heartsync/WAN2-1-fast-T2V-FusioniX
I found my issue and it was my own fault and nothing to do with the loras that you merged. :) It was another unrelated lora and sometimes two from the same creator downloaded from civitai doing it. :D The reward lora and perhaps the moviigen and realism lora to a tiny degree added to the "instagram look" but not nearly as much as I had previously thought.
Without any detail enhancing loras:
Just with the reward lora:
Just moviigen:
Just realism lora:
All enhancements:
All enhancements with one of my "faulty" loras at a strength of 1.0 added on top:
With the other lora at a strength of 1.0 :
With both at a strength of 1.0: