Lora trained on 189 images of Linda Cardellini on the SDXL 1.0 base model. The recommended Lora strength for most use cases is between 0.8-1.2. As always Adetailer is recommended for wider shots with the subject in the distance.
Description
FAQ
Comments (8)
Why did you remove the LoRA metadata? No trigger word? You didn't even train with captioned images? 'sigh'
I didnt remove anything intentionally. I normally dont use trigger words. It depends how good the dataset is. I have found that if the dataset is large, hgih quality and varied enough the captioning doesnt really add much quality (maybe a littel better flexibility but even that is not really a huge difference).
@steffangund Did you train text encoder? Which Optimizer did you use? Dynamic like DAdaptation Prodigy? Did you use sd-scripts,bmaltais/kohya_ss,Onetrainer,dreambooth,civitai? Also lots of stuff in the parameters are important for even being able to use it properly. (PS. Your captions are triggers words DS.)
@S1LV3RC01N I didnt train text encoder. I use Adafactor optimizer not prodigy. Kohya
@steffangund but you did captions on the dataset?
@S1LV3RC01N No captions
I love Linda Cardellini. Would you be able to make a Pony version of her? This LoRa doesn't work very well with it. But it works great with SDXL.
I will keep her in mind