Sup!
This is a slider LORA that helps with control for the closing of the eyes
In regular prompting, this lora goes from -1 to 1.
Negative values make the characters open their eyes while positives close them.
In my opinion though, the best way to use this lora is by inpainting, like in the examples shown. (with something like 0.6 Denoising strength).
The range when inpainting varis a bit, but not much. Maybe from -2 to 2.
You can get pretty decent accuracy. But most of the action happens in a single fraction.
So, you could have an eye closed at 0.5 and open at 0.6. Don't be afraid to go into really small changes on weight. Like 0.562. You can absolutely do that and you'll notice a difference.
I'll try my best to include prompts like that in the examples.
Mainly works for realistic models and haven't tried it much on anime or 2.5D.
Examples on how to control facial expressions with loras in this video I made. (hope it helps):
Trained with:
Description
Sup!
This is a slider LORA that helps with control for the closing of the eyes
In regular prompting, this lora goes from -1 to 1.
Negative values make the characters close their eyes while positives open them.
In my opinion though, the best way to use this lora is by inpainting, like in the examples shown. (with something like 0.6 Denoising strength).
The range when inpainting varis a bit, but not much. Maybe from -2 to 2.
You can get pretty decent accuracy. But most of the action happens in a single fraction.
So, you could have an eye closed at 0.5 and open at 0.6. Don't be afraid to go into really small changes on weight. Like 0.562. You can absolutely do that and you'll notice a difference.
I'll try my best to include prompts like that in the examples.
Mainly works for realistic models and haven't tried it much on anime or 2.5D.
Examples on how to control facial expressions with loras in this video I made. (hope it helps):
Trained with:
FAQ
Comments (9)
Wow this is impressive. I like how detailed you made it.
thank you!!
Can you upload a safetensor version?
I dont know how to make it a safetensor honestly. I'll look into it
I can't make this one to work in Fooocus, I added the PT file into the embeddings, but does nothing.
@sofiadidmendoza not 100% sure how Fooocus works, but this file is not an "embedding" but a "Lora". There should be a seperate folder called "Loras" or something, put the file in there
@Joppiesauz btw, found a way to train it in safetensors. But I lost the data on how tf I trained this. I'll try again with safetensors and upload it as a diff version when I can!
@N4Talent Thank you! Just so you know, you should be able to convert your already existing ckpt file to a safetensor without having to retrain it from the ground up. Ckpt files are just an older extension that is less safe than safetensors, thats why I'm hesitant to download it (not that I don't trust you, I've seen your youtube videos)
@Joppiesauz done!