Unreal Tournament 2004 - Facing Worlds - Sun Blade

This is my first Trained LoRA and also a Background...
Because I'am new to this training stuff I'am not sure which Epoch is the best. this is what CivitaAI's On-Site trainer recommended.

Thanks to Valstrix for helping me out and giving tips. And his Training Guide:
https://civitai.com/articles/3522/valstrixs-crash-course-guide-to-lora-and-lycoris-training

Anyways a lil story why i made this... Im obsessed with Unreal Tournament games, always loved the Egyptian Style of the Sun Blade faction.

I generated a lot of images with the On-Site CivitAI Generator with the Ankha character from Animal Crossings. I got a weakness for Egyptian stuff... but found the backgrounds pretty boring and wanted something unique and special... So what I did here is bringing the Capture The Flag map Facing Worlds 3 from Unreal Tournament 2004 into a LoRA so I could use that futuristic ancient egyptian style to my generated images.

I'am also planning to release the Sun Blade character DIVA. And also indoors backgrounds are coming.

I think I got the best results between 0.7 and 1.0. The images are generated with a 0.7 weight. facing3worlds is the tag that is used in all images, so this should be a trigger word.
If you are facing problems with random colored pictures or paintings on the walls please add the tags: paintings, pictures, picture frames to negative prompt. for some reason some of the hieroglyphs AI thinks they are paintings or picture frames.

After some feedback from NanashiAnon you should add the tag tanned in negative because the characters tend to become tanned.

It's trained on 72 ingame screenshots and these settings on the On-Site LoRA Trainer:

{
  "unetLR": 0.0003,
  "clipSkip": 2,
  "loraType": "lora",
  "keepTokens": 0,
  "networkDim": 16,
  "numRepeats": 6,
  "resolution": 1024,
  "lrScheduler": "cosine",
  "minSnrGamma": 0,
  "noiseOffset": 0.03,
  "targetSteps": 2556,
  "enableBucket": true,
  "networkAlpha": 16,
  "optimizerType": "Adafactor",
  "textEncoderLR": 0.0001,
  "maxTrainEpochs": 12,
  "shuffleCaption": false,
  "trainBatchSize": 2,
  "flipAugmentation": false,
  "lrSchedulerNumCycles": 3
}

Tags used in the dataset:

distant structures
entrance
fog
hieroglyphs
moon
obelisk
pillar
planet
plants
pond
pyramid
rock
sand
sky
skyline
stars
stone wall
tiles
tomb
tower
trees

The tag skyline generate most likely the a edge of the "world" with trees plants and the distant structures with some nice sky.