Marina Sirtis is an actress known for playing Deanna Troi in Star Trek: The Next Generation and Star Trek Picard.
Version 1.1 is based on 184 images of Deanna Troi, including her blue dress, purple jumpsuit, and various other dresses worn throughout the series. The Lora does have a tendency to add the communicator/badge and I will consider editing this out of some of the training pictures to reduce its impact on the results as SD does struggle to handle it.
I am open to suggestions on any improvements or other feedback you may have.
Description
Version 1.0 was based on a lora training guide, however the results were far below expectations. Version 1.1 increased the number of epochs from 3 to 20, which significantly improved the picture generation quality.
FAQ
Comments (5)
This model was really well trained! It produces some good results, but they're usually very underexposed and have a color shift. I color corrected and bumped up the exposure in Lightroom and the image looked a lot better.
My one suggestion would be to white balance the training data and bump up the exposure a bit. With those two corrections and your current training config, it should produce some awesome images!
I wanted to check out this LoRA to compare it to my own. Maybe my notes might be helpful.
The LoRA I trained does not include anything that isn't the blue uniform. There were 101 images and there were no tags. The LoRA settings were basically all defaults with AdamW, Linear, although I don't remember what number I used for the directory.
This model is really good for likeness, better than my own in fact, but doesn't flex at all. No matter what prompt I throw at it, it's usually just her sitting at a desk or on a bed.
I wish there was some sort of half-way point between ours because I think that would be ideal.
This is hands down the BEST LoRa on here. Imzadi has never looked sexier.
I am interested in an SDXL version of this lora
She's looking good!