Faces & Emotions – 18 Facial Angles and Expressions

This LoRA creates a complete horizontal 6×3 facial reference sheet containing eighteen portrait panels of the same character.

The character’s identity, appearance, clothing, rendering style, lighting, camera distance, and background remain consistent across the entire sheet.

Only the head angle, eye-gaze direction, and facial expression change.

Sheet Layout

Left Side – Head Angles and Gaze Directions

Top row:

1. Looking up-left

2. Looking straight up

3. Looking up-right

Middle row:

1. Exact left profile

2. Straight frontal view

3. Exact right profile

Bottom row:

1. Looking down-left

2. Looking straight down

3. Looking down-right

Right Side – Facial Expressions

Top row:

1. Neutral

2. Open joyful laughter

3. Clearly angry

Middle row:

1. Visibly sad

2. Startled surprise

3. Playful and mischievous with the tongue sticking out

Bottom row:

1. Gentle smile

2. Shy or embarrassed

3. Tense or uneasy

Important Resolution Information

The most reliable base resolution is exactly:

1672 × 944 pixels

This resolution reproduces the intended 6×3 arrangement most consistently.

The LoRA has learned not only the expressions and viewing angles, but also the spatial structure of the complete sheet. Changing the width-to-height relationship can therefore affect the layout.

A canvas that is too wide may cause the model to generate additional rows, columns, or portrait panels.

A canvas that is too narrow may cause panels to disappear, become compressed, or merge incorrectly.

You can increase the resolution, but the width and height should always be scaled proportionally.

Recommended resolution steps:

1× Base Resolution
1672 × 944

2× Resolution
3344 × 1888

3× Resolution
5016 × 2832

4× Resolution
6688 × 3776

Higher resolutions require significantly more VRAM and generation time.

For the first test, I strongly recommend using exactly 1672 × 944 pixels.

You can increase the megapixel setting afterward, provided the same proportions are preserved.

Generating below the base resolution is not recommended because every individual portrait panel receives less image information, which may reduce facial quality and layout stability.

Activation Tag

The activation tag is:

FEPOSEGRID

Place it at the very beginning of the positive prompt.

Recommended Activation Sentence

For the most reliable results, use the complete activation sentence instead of only using the activation tag:

FEPOSEGRID, a complete horizontal 6 by 3 facial pose and expression reference sheet with eighteen equal square portrait panels and exactly the same single character in every panel. Identity, sex or gender presentation, age, species or visible human appearance, skin or fur color, facial anatomy, eye color, hair or mane, facial hair, makeup, markings, clothing, accessories, rendering style, lighting, camera distance, and background remain unchanged. Only head angle, eye-gaze direction, and facial expression vary. The left three columns use this fixed sequence: top row looking up-left, looking straight up, looking up-right; middle row exact left profile, straight frontal view, exact right profile; bottom row looking down-left, looking straight down, looking down-right. The right three columns use this fixed sequence: top row neutral, open joyful laughter, clearly angry; middle row visibly sad, startled surprise, playful mischievous tongue-out expression; bottom row gentle smile, shy or embarrassed, tense or uneasy.

How to Prompt the LoRA

The positive prompt should contain two main sections.

First, use the complete activation sentence.

After one blank line, describe the character you want to generate in detail.

Recommended prompt structure:

[Complete Activation Sentence]

All eighteen panels depict the same [detailed character description].

Your character description should clearly define:

• Age and gender presentation

• Ethnicity, species, or visible appearance

• Skin, fur, scales, or surface material

• Face shape

• Eye shape and eye color

• Eyebrows

• Nose

• Lips, muzzle, beak, or mouth anatomy

• Hair, mane, fur, horns, or head features

• Facial hair or makeup

• Clothing

• Accessories

• Rendering style

• Lighting

• Background

Avoid adding separate poses or additional emotions to the character description because the activation sentence already controls the full layout.

Example Prompt

FEPOSEGRID, a complete horizontal 6 by 3 facial pose and expression reference sheet with eighteen equal square portrait panels and exactly the same single character in every panel. Identity, sex or gender presentation, age, species or visible human appearance, skin or fur color, facial anatomy, eye color, hair or mane, facial hair, makeup, markings, clothing, accessories, rendering style, lighting, camera distance, and background remain unchanged. Only head angle, eye-gaze direction, and facial expression vary. The left three columns use this fixed sequence: top row looking up-left, looking straight up, looking up-right; middle row exact left profile, straight frontal view, exact right profile; bottom row looking down-left, looking straight down, looking down-right. The right three columns use this fixed sequence: top row neutral, open joyful laughter, clearly angry; middle row visibly sad, startled surprise, playful mischievous tongue-out expression; bottom row gentle smile, shy or embarrassed, tense or uneasy.

All eighteen panels depict the same photorealistic adult woman in her early thirties with warm medium-brown skin, a softly oval face, pronounced cheekbones, dark almond-shaped eyes, thick naturally curved eyebrows, a straight nose, full lips, and shoulder-length tightly curled black hair. She wears the same dark green high-neck top and small gold hoop earrings. The background is a smooth light-gray studio backdrop with soft balanced portrait lighting.

Tested Character Types

The LoRA has been tested with a wide range of subjects, including:

• Photorealistic men and women

• Anime characters

• Furry characters

• Fantasy creatures

• Monsters

• Androids and robots

• Aliens

• Elves, orcs, demons, and other non-human characters

The preview images intentionally use very different characters and visual styles to demonstrate the flexibility of the LoRA.

The complete prompts used for the example images are included below the corresponding images.

Created by Astroburner