I noticed that Stable Diffusion had no concept of the regency period, the period of British history made popular by Jane Austen novels and her many imitators. Using film stills from various adaptations, I made this LORA.
In the training data, I used tags like "regency period dress," "regency period fashion," and "regency period outfit" to describe clothing, so those tags will likely work best.
The model is biased towards photos, so you may want to use tags like "film still" or "photo" in the negative prompt.
I've been using strength between 0.5-0.7 and it seems to work pretty well at that strength.
Description
Initial version trained on film stills and photographs
FAQ
Comments (9)
"albedobaseXL_v20", why upload it as a SD1.5 LoRA?
I thought I'd selected SDXL when I published it, but when I checked in on the model a later saw it was labelled SD1.5. It's fixed now.
@masilverhammer182 I don't suppose you could train a 1.5 version as well? Machine doesn't do too well with SDXL models.
@Trefolder Yeah, I could do that. I think all I'd need to do is resize the images to 512x512. I used CivitAI's lora trainer, so it'll cost some buzz but that's pretty easy to earn.
@masilverhammer182 How much buzz would you say? I'd be happy to donate my meager amount to the effort to at least offset things.
@Trefolder I'd appreciate that! It costs 500 buzz to train a Lora, so if you donated 100-200 buzz that'd be awesome.
@masilverhammer182 Done! or at least I think it's been done? Let me know!
@Trefolder So I can train on standard SD1.5, anime, semi-realistic and realistic. Do you have a preference?
EDIT: all my training images are film stills or backstage photos so I'll do it on standard SD1.5 for greater flexibility.
@masilverhammer182 Standard SD 1.5 should be fine, if it doesn't work I can see about getting another 500 for some anime training maybe.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.