Amy Adams Reviews

Somehow felt the need to write a review about this one, because I’ve recently experimented a lot with SDXL loras and faced some of the issues that the creator of this model probably encountered as well.

If you feel the need to create a LoRA for something that several people have done before, it is always good to bring something unique to the table. In this case I assume that was limiting the training to a certain role of the actress. In this case it would have been better to do a character LoRA and not market it as a celebrity one, but that is a personal choice.

As it’s a celebrity LoRA and one for a popular actress, I can compare it to an existing one here. Let’s take the Amy Adams SDXL from razzz for example. Doesn’t look quite fair, because it’s only half the size of this 1.9 GB monster, but anyway.

I’ve taken the liberty to do comparison images, much more fun that way. The LoRA from Tophness is actually quite good if you go more for a painting style of a portrait. It also captures the screenshot vibe that is often lacking sharpness (due to moving subjects, as it’s a movie and not a still photography source) with a hint of motion blur in it.

It’s lacking flexibility, because the hair color of the actress was not tagged, so she will always have red hair (unless you use weighted prompts and/or combine it with negative prompts). The LoRA from razzz is much better in that regard. Then you can argue that it’s because she has red hair in the movie that was used for training, but it’s a celebrity LoRA and not a character one, so it should still provide some flexibility.

If you prefer the certain style that this LoRA offers, it might be worth to download this model (a pruned SD 1.5 checkpoint is almost the same size). I personally know which one I’m going to delete later.