Likeness of Emma Stone :)
"Natural red hair" will help you avoid a vibrant dyed look. I don't mention skin tone or anything because normally the text encoder is smarter than me.
Description
FAQ
Comments (11)
Hi. You given traininginformation as epoch 85, but would be interesting the count of training images to. Maybe the training dataset it self?
Sure, one sec and I'll upload them. I want to say it was around 30 images
@direct22202 Thanks so much. So you didn't use any text files?
@Clocksmith I also need to know this!
@thefoodmage just trigger word! Save it just like you would a full caption. All my captions only have the trigger word.
@Clocksmith only the trigger word in each text file. No description or anything!
@direct22202 I'm happy to hear that. It is one of the best lora's I've used, incredibly expressive. It makes me think other people are ruining their loras by putting excessive prompts in the training files.
@Clocksmith thank you very much! I think a diverse dataset with different angles is key too. What I keep reminding myself is just because I think a picture looks good or it's high resolutions means that it's going to make a good dataset selection.
@Clocksmith but yes I definitely feel that my previous experience training personal loras were very limited by my own use of auto captions/GPT/etc.
😍 👍🏻
Thank you guys for the buzz and the reactions! I feel like this one turned out so well.