A character LoRA featuring four members of a jungle Amazon tribe: the Queen, the Shaman, the First Warrior and the First Huntress. All four are trained into a single LoRA and can be used on their own or together in one image.

Characters & Trigger Words

•      qu33n_amz – The Queen: Black band of war paint across the eyes, feather crown, large feathered earrings, tribal tattoos on arms and legs, black sun tattoo around the navel

•      sh4m4n_amz – The Shaman: Long pale blonde hair, wolf head headdress, white clay paint (dots on the cheekbones, three lines on the chin)

•      w4rrior_amz – The First Warrior: Muscular build, long black hair with a shaved left side and an angular tribal tattoo on it, bold angular tattoos on arms and thighs

•      huntr3ss_amz – The First Huntress: Athletic, lean build, copper-red hair in one thick braid, green stripes of paint on both cheeks, vine-like tattoos

The signature features are built into the trigger words. You don't need to describe hair, face paint, tattoos or headwear.

Clothing is modular

Clothing is not baked into the characters. This means you can dress them in anything you like. It also means: if you don't describe any clothing, the characters will usually appear unclothed. Please keep this in mind when prompting.

To get their standard outfits, use these descriptions:

Queen

wearing a bone necklace around her neck and a crude thick dark brown leather loincloth, held in place by a belt decorated with animal teeth around her waist, crude hand-made boots made out of fur and rope on her feet

Shaman

wearing a crude loincloth made of wolf pelt, held in place by a crude hand-made wide belt made of brown leather with a crow skull as belt buckle, crude hand-made boots made out of fur and rope on her feet

First Warrior

wearing a sleeveless cuirass of hardened brown leather with bronze fittings, a short skirt of leather strips, bronze bracers on both forearms and hand-made leather sandals laced up to the knee, a bronze short sword in a scabbard at her left hip, a round wooden shield with a bronze rim on her left arm

First Huntress

wearing a brown leather tunic trimmed with pale sheepskin, a short green-brown leather wrap skirt and soft leather boots wrapped with pale sheepskin, a quiver of arrows slung across her back, holding a longbow in her left hand

Prompting

Single character

huntr3ss_amz, woman, full body shot, front view, standing, wearing [outfit], on a rocky beach, overcast daylight

Two characters

two women standing side by side, on the left w4rrior_amz, woman, on the right huntr3ss_amz, woman, upper body shot, looking at the camera, jungle village, daylight

Three characters

three women standing side by side, on the left qu33n_amz, woman, in the middle huntr3ss_amz, woman, on the right sh4m4n_amz, woman, upper body shot, looking at the camera, jungle village, daylight

Tips

•      Always follow the trigger word with woman.

•      Describe what you want to see, not what you don't want. Krea 2 Turbo tends to add things that are mentioned in negated form (e.g. “no headdress”).

•      Recommended settings: LoRA strength 1.0, about 8–10 steps, CFG 1.0, sampler euler / scheduler simple.

Known limitations

•      Multiple characters: Two- and three-character images work in most seeds. Occasionally a feature still jumps from one character to another – most often from the Queen, who has the strongest features (sun tattoo, crown). If that happens, simply try another seed. Images with all four characters together have not been tested systematically.

•      Fur in prompts: Words like fur or sheepskin in a prompt (for example for boots) can add pieces of fur elsewhere on the body. If a character should wear only boots, try a few seeds or describe the boots with other materials.

•      Generic women: With the LoRA loaded, the word woman without a trigger tends to produce an Amazon as well, and background figures can pick up traits of the main character. Lowering the LoRA strength reduces this, but also weakens the characters.

•      Small details (face paint, fine tattoos) can get lost in wide shots. Closer framing or a detailer pass helps.

Training details

•      Base model: Krea 2 Turbo

•      Trained with ai-toolkit, rank 128, resolution 1024

•      Dataset: ~260 synthetic images – single-character images, two-character images (clothed and unclothed) and group scenes with one character among other Amazons – captioned with trigger words, framing, outfit, setting and lighting

•      Selected checkpoint: step 11,750 of 13,000

Created by apehead.