Accuracy of the characters when viewed from a distance.
Hey!
First off, congrats on this LoRA! It's incredibly useful. Thanks so much for making it!
I have a question and a suggestion for future versions, if you're planning to keep developing it.
Here's how I currently use it: I take my 3D renders from Blender, use AI to overpaint them and improve the lighting and textures, and then use your LoRA to transfer that new look onto my original 3D animation.
The problem is that when characters are further away from the camera, they tend to lose facial details during the process.
I'm not sure exactly how your LoRA handles transferring the new look onto the original animation. Does it internally estimate things like Z-depth, normal maps, position maps, or something similar?
If so, here's my suggestion: Would it be possible to add an option that lets us provide those auxiliary passes manually?
Since I'm working with my own 3D scenes, I can render all those passes directly from Blender, so they'd be accurate rather than estimated. I'm guessing this could help preserve facial details and make the appearance transfer more faithful to the original animation, especially for characters in the distance.
Just an idea! Thanks again for your amazing work!