Sign In

Dana Scully Character LoRA SD

93
1.1k
18
Verified:
SafeTensor
Type
LoRA
Stats
438
0
Reviews
Published
Feb 24, 2024
Base Model
SD 1.5
Usage Tips
Clip Skip: 1
Strength: 0.8
Trigger Words
agentscully
Hash
AutoV2
6085CCE15F
dbst17's Avatar
dbst17

Dana Scully

A LoRA for Dana Scully, a fictional character from the television series The X-Files, portrayed by Gillian Anderson.


This is based on the earlier episodes, which I still consider the best.

There are already a few models for this specific character available here, just wanted to add one more. Most checkpoints already have an implementation for this, so the output can be quite different, depending on the model you use. It can also be very hard to change her hair color or hair style, because of the baked in preferences for the character. But after all it's a character model, so it comes with default parameters.


This is the SD Version based on the same data as my SDXL one. I recommend to use Hires.fix for optimal results. If you want to work with lower resolutions: ADetailer will do a fine job.

Version 2

I recently did a bounty for the Dana Scully character for Season 5-7 and I now decided to combine the training dataset from Version 1 with the additional images I gathered from the later Seasons of The X-Files.

Version 2 covers a wider range and also has new tagging for the images. I also went for a new trigger word, which allows it to not mess with a trained character in the checkpoints and lead to better results. It also eliminates the need for the stronger version. This one worked pretty well with comic and semi-real models.

Here is a direct comparison of the versions:

Version 1 still works very nicely, while Version 2 goes for a little more mature look. Mainly a matter of preference.

Version 1

There are two versions. The normal one is for models that already have an understanding of the character (most western models) and a strong version that don't (several asian checkpoints).

Select the one that suits you needs, they are based on the same data, the stronger version has received more epochs but will appear overtrained for models that already know the character.