Mechanicum: Warhammer 40K (SDXL)

Name: Mechanicum: Warhammer 40K (SDXL)
Rating: 5 (29 reviews)
Author: ndimensional

736

4.1k

Updated: Apr 15, 2024

concept

scifi warhammer sci-fi 40k woman

Images hidden due to mature content settings

Verified: 10 months ago

SafeTensor

Details

Type	LoRA
Stats	736 4,081
Reviews	Very Positive (92)
Published	Jan 8, 2024
Base Model	SDXL 1.0
Usage Tips	Clip Skip: 1
Hash	AutoV2 10A1633169

1 File

About this version

default creator card background decoration

ndimensional

Images hidden due to mature content settings

"The eyes of the Omnissiah are ever upon us." - Plutonic, Warhammer 40k: Mechanicus.

This is the first (of many) LoRAs I plan to release over next few months as I work on my ambitious Warhammer 40k full-parameter fine-tune.

40K Series: Necron, Drukhari, Astartes, Sisters of Battle, Death Korps, T'au Empire, Aeldari, Oldhammer

Mechanicus

This particular LoRA focuses on the Adeptus Mechanicus.

Things to know:

The only special tokens needed are tech priest or Adeptus Mechanicus. Although these tokens are not strictly required.
I found that using a LoRA weight of 1 works best with SDXL.
This dataset was not strictly created to be a LoRA. The data comes from a much larger 40k dataset I'm putting together. As a result, this LoRA may have some quirks.
This LoRA will only be as good at NSFW as your base model.
As of writing (1-8-24): An update will be available shortly. I forgot to train some of the subdirs of the Adeptus Mechanicus dataset that featured things like, servo-skulls, skitarii rangers, skitarii vanguard, kastelan robots, imperial robots, and cyber-constructs. Once this error is corrected, the LoRA should be more comprehensive.
An SD1.5 version of this LoRA will be available shortly.

Warhammer 40k: Diffusio Schisma

As mentioned early, this LoRA comes from part of a dataset for a much, much larger model.

The full model will cover nearly all of the Warhammer 40k universe: races, species, daemons, factions, sub-factions, terrain, vehicles, weapons, ect..

How?

Thanks to the advancements in unified embedding spaces with multimodal LLMs, it is now possible create accurate, highly detailed, guided image captions for large databases with less manual captioning.

That being said, it's not perfect and requires a lot of time and programming to pull this off.

Knowing the full model is still a little ways off, I will be periodically releasing LoRAs for both SDXL and SD1.5 that are trained using portions of the full dataset.

You can find a more comprehensive list, more information, and track the progress on this model in the discord thread bellow.

Note: I will occasionally put out polls for which part of the dataset to make a LoRA from next, on the discord thread linked bellow.