Natively-trained LoRA of UK actress Gabrielle Drake, a popular face on TV and in movies in the UK in the 1970s. She is perhaps best-known for her purple-haired portrayal of 'Lt. Ellis' in the 1970 Gerry Anderson sci-fi TV show UFO.

Uses 105 ref images and 1989 (real) reg images, trained for 3 epochs of 3000 steps, at a batch size of 1, using BF16 on a 3090, with a training time of three hours. Data was hand-annotated, aided by suggestions from WD14, and further curated in BooruDatasetTagManager.

Though this uses the same limited image data as my previous checkpoint model of Drake, it is far better, more accurate, and more disentangled. However, the majority of the data had to be colorized, and it shows in the output.

There is almost zero good-quality data available for this personality. So you will need to get out and push with this model. It will do well on a low CFG (around 3.5-5) and a modest/moderate denoising strength in inpaint.

What, No UFO..?

Fans of Lt. Ellis from Gerry Anderson's UFO may be disappointed to find that there is no 'moon maiden' data in this model - the 'Ellis' look is just too extreme, and would overwrite all the other data. So it would be more suitable for a dedicated model of its own.

It's a shame UFO material could not be used, since HD material is available; but Drake only appears twice, and very briefly, without the over-the-top costume, wig, and make-up - in the episodes 'Close-Up' and 'Computer Affair'. Sadly, those quick clips are not much use for training.

Age Adjustment

The model covers a long period, from the actress's late twenties to her late thirties; use 'younger' or 'older' to affect appearance.

Usage

I generally use these models in complex workflows, inpainting faces after initial T2Is, and using ControlNet extensively. So if you're hoping for one-click prompt magic, my models aren't data-curated with this in mind, but rather as tools for traditional workflows that use Photoshop and other older methods.