PLEASE NOTE - ALL MY CELEBRITY LORAS WILL BE DELETED (BY MYSELF) BY THE END OF JUNE 2024.
This model was trained at 512px x ??0px. Consequently it rarely 'doubles up' when using the popular 512px x 768px format.
A LoRA of 1960s/70s US TV actress and Bewitched star Elizabeth Montgomery, using 72 ref images and 1980 (real) reg images, for 27,000 steps over three hours on a 3090 GPU. Trained from scratch in Kohya, not extracted from a DreamBooth.
This replaces the checkpoint I did previously, and is more flexible than that version.
In choosing a checkpoint from those saved during the training, I've tried to strike a balance between flexibility and detail, since the earlier checkpoints are fabulously flexible but the later ones (the ones that are a bit over-trained and overcooked) have the best detail. As is usual with these LoRAs, 5 epochs was the compromise here.
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.
This model is very likely to produce ????? renderings unless counter-prompted.
Description
This model was trained at 512px x ??0px.
Trained on 77 ref images and 1980 reg images over 5 epochs on a 3090, for 27,000 steps over four hours, with BF16 and sub-concept ('lizmontface' - may give better detail in full-body pictures if used as an extra prompt, i.e., 'lizmont woman with lizmontface').