The Lord of the Rings style LoRA
Making models can be expensive. Do you like what I do? Consider supporting me on Patreon 🅿️ or feel free to buy me a coffee ☕
Trained on AbsoluteReality 1.6, triggers with Lord of the Rings. but be careful because "lord" is a male word. This lora helps to increase the realism and do incredible medieval/fantasy looking stuff. I mostly made this to make realistic D&D characters for my campaign.
Careful because "lord" might produce male people on some models. lotr should work fine too as it should have a similar encoding on most models.
I had a lot of fun using it. Should work around weight 1 on AR (all versions, but mostly 1.6).
Some examples are too "sexy". You need to register to see them.
How to use LoRA's in auto1111:
Update webui (use
git pulllike here or redownload it)Copy the file to
stable-diffusion-webui/models/loraSelect your LoRA like in this video
Make sure to change the weight (by default it's
:1which is usually too high)
Description
FAQ
Comments (11)
Since we kind of have to work around the parameters and quirks of using stable diffusion and natural language versus their interpretation of language... I'm glad you clarified the thing about "Lord" being maybe a male thing. Would a fix for this be as simple as changing the triggers or handling to something like "LoTR" instead of something with "Lord"?
Please do use lotr for the trigger word next version, Lykon. It's a well-known abbreviation and SD shouldn't know it.
@ptdurnvcoedqegsdnw should have roughly the same encoding, so you can probably use it already with this version.
the point of the trigger is that's a finetune over the base model text encoder. Quite likely Absolute Reality understands "lord of the rings" pretty well enough, so it didn't get a strong finetune. However "lotr" should have a similar encoding already. The fact is that I can't predict how most models will behave with their text encoders. Even using "lotr" as trigger during the training might have almost no effect since AR understands it well as base. It is also not really required, the model works well even without "from lord of the rings" ;)
In general I can't predict how this will behave on all models, regardless of the word I choose. If you go outside of the suggested settings you have to be prepared to fiddle with it. I can test only some settings before release, can't test 1 billion models and combinations :D
amazing once again, though I've noticed a issue with this and the GOT/80s/burton models is they tend to generate faces similar to there franchise actors, do you know of anyway to circumvente it? also discovered GOT strangely does very nice sci-fi images great for planetary romance/medieval sci-fi images makes me wonder if a Lucas model would work well?
It was pretty hard to have faces not look exactly like them, which is something I do on purpose to avoid legal issues.
Given that casts are limited, it can't really be helped to get similarities. I guess mixing them up and fiddling with weights, as well as trying to condition with prompts is the only way.
@Lykon ah fair, I'll give that a try, though I've noticed it's best at making gondor or rohan style characters, easterling or mordor characters are difficult, did you only use screenshots of the heroes in the data?
@stygianwizard42 no, but I guess it's CLIP knowing humans better than monsters :)
@Lykon no I mean even trying to get human style characters who look like they're on Mordor's or Isngard's side is difficult
@stygianwizard42 you can probably make an embedding for that quite easily. You don't see Mordor humans often in the movies.
Good one as always, author-sensei. You should've been the showrunner for that Amazon LotR. It would've at least not sucked.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.