Type | |
Stats | 553 2,229 |
Reviews | (82) |
Published | Aug 30, 2023 |
Base Model | |
Training | Steps: 4,025 Epochs: 150 |
Usage Tips | Clip Skip: 2 |
Trigger Words | cav_rdrcolter dilapidated cabin frozen stream winter |
Hash | AutoV2 B32592B0C8 |
About:
The snowbound town of Colter, in northern Ambarino, is an abandoned settlement in Red Dead Redemption 2. Colter provides the setting for the first chapter of the game, and it's here that I went on a photo gathering expedition.
The area depicts a number of collapsed and dilapidated cabins buried in the snow. A frozen stream runs through the forest behind the town, and high rocky mountains ring the horizon. Trees in the area are thin, often broken, and beaten down by the harsh weather, lending further atmosphere to this inhospitable place.
This LoRA has been created from 161 images of the town and surrounding landscape during the game's opening chapter. It's suitable for a range of styles, from anime and cartoon, all the way through to photo realistic.
Use:
At a full strength weight of 1.0, it will produce images very close to the PC version of the game at high detail levels. The opening chapter takes place during a white-out snowstorm, so generated images may lack saturation.
Weights between 0.7 - 1.0 are suitable for animation styles
Weights between 0.6 - 0.8 will get you into photo-realism territory with an appropriate checkpoint. Dial the weight up and down as needed.
While realism increases with lower weights, the grungy nature of the architecture will decrease, so take that into consideration.
Trigger Words: cav_rdrcolter is the main trigger. You can also use place (the training class) dilapidated cabin snow winter frozen stream mountain log cabin stone cabin or broken fence. I'll be uploading a wide range of example images with full metadata, so take a look at those for some suggested starting prompts.
What's Next?
This is my first LoRA, and it took me over a dozen different iterations to get to a point that I felt worthy of a 1.0 designation. After working with it a bunch, I think there's room for improvement still. Primarily, I want to return to the area later in the game, when the weather is better, to get training images that aren't washed out from the snowstorm. There are also many other distinct and architecturally interesting locations throughout the game that I want to make additional LoRAs from.