quality work once again, how many images in the dataset and which resolution(s) do you use for training? Your work is absolutly insane.

exactly how many of your training images were in front of one of those award show walls covered in logos?
I ask because >50% of gens I get with simple prompts are coming out in front of distorted versions of those award show walls (obv because SD can't render the text properly), but I recently trained a LORA with 200+ images nearly half of which were also in front of those damned award show walls and that LORA never gens the walls. Which frankly startled me.. I just threw my images into a single folder with zero captions and set the training loose on it. ;)

https://civitai.com/models/143480/aulii-cravalho-lora