##### V2 is here! #####

v2 is the same dataset, but with better resolution, bucket training, better LoRA rank and better captions for materials.

This LoRA is trained on a large variety of tribal, nordic and shamanic outfits. It includes clothing made from natural and handcrafted materials such as leather, fur, hide, linen, and woven fabrics. Many outfits feature decorative elements like feathers, beads, bones, metal ornaments, ropes, and wood. Common accessories are necklaces, amulets, arm cuffs, belts, and staffs with charms or skull details. Hairstyles often include braids, dreadlocks, or loose hair decorated with beads, cords, and feathers. Headgear is a major element in this LoRA, ranging from antler crowns, horned helmets, hoods, and feathered headdresses to circlets and bone masks. The LoRA also covers ritual tattoos, tribal body paint, and symbolic markings in black, white, red, or blue tones. Most images are set in natural environments such as forests, snow, rocks, or fog, emphasizing earthy colors and organic materials.

The LoRA was trained on the default flux1-dev model.

Dataset and training data:

• 204 images

• 5 steps per image

• 10 epochs

• bucks with 1152px max size, no auto scaling

• cosine scheduler with 0.2 warmup and 0.7 decay

• Captioning was done with OpenAI

• alpha == dim: 32

If you use it with other loras, for example a character lora, try to lower the weight - I had good results at around 0.5 and lower.