Bite Me: Umbrium Regional Mouths

作者：umbrium2835

Regional Mouths adds extra mouths to your character’s body, with separate triggers for nine regions. It is designed for imaginative character illustration, with prompt controls for placing an additional mouth on a selected region. This is one combined LoRA. Choose the region in your prompt, then describe whether the body mouth is closed, open or showing its tongue, and whether its teeth are flat or sharp. How to use Start with one regional trigger:

• mouthstyle arm

• mouthstyle back

• mouthstyle chest

• mouthstyle face

• mouthstyle foot

• mouthstyle hand

• mouthstyle leg

• mouthstyle neck

• mouthstyle stomach

Add the body-mouth controls as well. A tested starting combination is body mouth open, body mouth flat teeth. The region trigger alone can produce a weaker or missing effect.

• Mouth state: body mouth closed, body mouth open or body mouth tongue out.

• Teeth: body mouth flat teeth or body mouth sharp teeth for open mouths.

• Paired limbs: two body mouths was included in a small number of training examples; treat it as experimental.

Describe the character, clothing, pose, camera angle and setting normally. Keep the requested region visible and large enough in the image. For a face placement, ask for an extra mouth separate from the ordinary mouth. Example prompt mouthstyle arm, body mouth open, body mouth flat teeth, 1boy, adult man, solo, short hair, sleeveless shirt, upper body, arm visible, outdoors, overcast Illustrious v1 settings

• Base family: Illustrious. Use the file with an Illustrious-compatible checkpoint.

• Starting strength: 1.0 for both model and CLIP. Reduce the strength if it changes the character or scene too strongly.

• Use your usual checkpoint-specific generation settings. The regional trigger and body-mouth controls are the important additions.

Training and testing Trained on 355 individually reviewed images with regional captions: 40 each for arm, back, chest, face, leg, neck and stomach, 39 for hands and 36 for feet. The dataset includes male and female characters, single figures and pairs, and several illustration and realistic styles. The Illustrious version was trained from Illustrious XL v2.0 for 2,000 optimizer updates, with rank 32 / alpha 16 and model learning rate 0.0005. Testing compared Off, 1,000 and 2,000 steps across two prompts per region: 54 images total. Known limitations — Illustrious v1 The final checkpoint placed an extra mouth in the requested region in 14 of the 18 test scenes. That measures placement in a small test set, not clean anatomy or a guaranteed success rate.

• Chest, back, stomach, leg and neck placement worked in both tested scenes for each region.

• Hand placement failed in both tested scenes; mouths appeared elsewhere instead. Feet and extra facial mouths were also inconsistent.

• Mouths may appear on nearby regions, stretch along a limb, or alter the ordinary facial mouth. Check placement and anatomy carefully.

• Style, apparent age, clothing, pose and framing can change. The LoRA does not preserve every detail of the base image.