NOTE: I am so exhausted trying to create women's apparel that is modest enough to not get flagged by civit's filters (1cm of skin below the middle of the throat is enough to trigger it sometimes, and at other times it seems to be turned off entirely), that I have decided henceforth to only post face images.

INSTRUCTIONS (IMPORTANT!)

1. Use a LoRA strength between 0.1 and 0.15. Anything more than that and you will get absolute garbage (see 'the three-times trained method' below).

2. You may need to use negative prompts about male characteristics such as '1boy', and 'man', and 'ripped', etc (see 'This model is an interesting failure' below).

3. The addition of the word 'female' and traditionally female-related positive prompts can help to stop Stable Diffusion trying to interpret Jamie Lee Curtis as a man (see 'This model is an interesting failure' below).

4. If you want to reproduce the classic look of Jamie Lee Curtis, i.e., the short spiky hair of Trading Places and other mid-80s movies that she did, you will have to emphasise this, perhaps multiple times in the prompts, because the use of female emphasis in prompts is likely to cause Stable Diffusion to make Curtis's hair longer, and in general to conform her to its own idea of female beauty.

The 'Three-Times Trained' Method

A Kohya LoRA of American actress Jamie Lee Curtis, covering the period 1983-1985. It was trained THREE TIMES on 130 images, with 1980 reg images. It was trained for two hours on a 3090 GPU, for four epochs for each training set, and the three resulting models merged in Kohya.

This LoRA uses a new technique first shared on Reddit in late September 2023, by the user shootthesound. Please see the above link for details, but the long and short of it is that you create two versions of the same training data (one portrait and one square, i.e., for instance, 512x512 and 512x768), and train a LoRA for each of them.

You then pick the best trained checkpoint from each and merge them in Kohya at 100% strength each. See the original post for comments from a machine learning expert as to why this massively improves the quality of the LoRA, but suffice to say that the merged LoRA now has the best of both worlds.

The fact that they are merged at 100% each is why you need to use LoRAs made with this technique at around 0.4 strength, because technically the two LoRAs represent a 200% strength!

There are several benefits with this approach:

1. Faces are much more detailed, even when they are small in frame.

2. The overall quality is extraordinarily magnified.

3. You can ramp up the CFG almost to the end of the scale before there is any degradation of quality, which means your prompt instructions can be followed without sacrificing quality.

4. The resulting LoRA is incredibly disentangled, and can adopt poses and characteristics present in the LAION dataset that do not exist anywhere in the training material that you used for the LoRA.

And there's more besides this - but try it for yourself, and see for yourself.

But...

This model is only an interesting failure

Despite being recognized as one of the hottest women of the 1980s, I've noticed that the fairly androgynous face of Jamie Lee Curtis is frequently identified as either male or transvestite by AI systems - and most especially by Stable Diffusion-related systems such as CLIP and BLIP, for which Curtis's unusually saturnine face tends to trigger a 'male' association.

This means that no matter how good the data in the training set is, for pretty much any SD model you make from images of her (this is my fifth and probably final attempt), you will have to constantly use 'male'-related negative prompts (such as '1boy', and 'man'), in order to stop Stable Diffusion from reducing Curtis's female characteristics, or from adding male characteristics such as flat chests and ripped muscles.

Though I think this model provides a better resemblance than others of Jamie Lee Curtis that are available at the time I write this, I believe that the problem lies with the trained assumptions and stereotypical generalizations of Stable Diffusion's core models.

This model is very likely to produce NSFW and nude renderings unless counter-prompted.