Natasha Vale

117 training images, 23,000 steps, 20 vectors per token, learning rate of 0.0015.

I am experimenting with alternative 'recipes' for TIs, this is my standard one. I'm trying to speed up the process and improve flexibility.

SD does seem to like giving her the 'Rachel' rough bob haircut. To be fair, that was a look that Natasha had a lot - we're talking the late 90's, early 00s after all! It's possible to give her shorter or longer hair with strongly-weighted prompts.

Many of the sample images are the same seed, with prompt modified only for clothing, but trying it in different models.

N.B. There were a lot of NSFW images in the training data. Even with strongly-weighted SFW/NSFW in the +ve/-ve prompts it can still throw out the occasional bit of nudity. If this bothers you, I suggest joining a monastery / nunnery.