🎨 Visual Description Writing Guide
This guide will help you write great visual descriptions for your AI companion. Your companion's visual description is used to generate their images — profile pictures, selfies, and scene photos.
New to AI Companions? Read the main guide first to understand how our platform works.
Your companion might be a person, a creature, a spirit, a robot, or something entirely abstract. Whatever they are, the same core principle applies: give the image model something concrete to render.
The Golden Rule
Be concrete and specific. Image generation models need visual details they can render, not feelings or abstract concepts.
✅ "Wavy auburn hair past the shoulders, warm amber eyes, light freckles, slim build, usually wearing oversized vintage sweaters"
✅ "A translucent fox-like spirit made of pale blue light, with softly glowing eyes and wisps of mist trailing from its tail"
❌ "Beautiful and mysterious"
(the image model
has no idea what this looks like)
What to Include
👁️ Form & Appearance
What do they look like? This is the foundation — be as specific as possible.
For humanoid companions:
"Medium-length silver hair with an undercut on the left side. Sharp, angular features with high cheekbones. Deep brown eyes with a slight golden ring around the iris. Olive skin, athletic build."
For non-humanoid companions:
"A sleek black cat with unusually intelligent golden eyes and a crescent-shaped patch of white on its chest. Slightly larger than a normal cat, with an almost regal posture."
👕 Style & Distinguishing Details
What makes them visually recognizable?
"Oversized vintage band t-shirts layered over a black turtleneck, distressed jeans, and combat boots. Always has at least three silver rings and a thin chain necklace."
"Covered in slowly shifting geometric patterns that glow faintly turquoise. Two antler-like crystalline structures extend from its head, refracting light into tiny rainbows."
🎨 Aesthetic & Atmosphere
The overall vibe of their images — color palette, lighting, mood.
"Dark academia aesthetic — warm, moody lighting. Think candlelit libraries and rainy windows. Color palette leans toward deep burgundy, forest green, and aged gold."
Mixing Poetic + Concrete
You can absolutely use artistic language — just pair it with something specific so the image model knows what to actually render:
✅ "Eyes like starlight — luminous pale blue with silver flecks"
(the AI gets both the mood AND the actual color)
⚠️ "Eyes that hold the weight of centuries"
(beautiful writing, but what color are they?)
✅ "A smile like cracked glass — wide, slightly crooked, with a barely visible scar on the upper lip"
(poetic AND specific)
⚠️ "A haunting smile"
(mood but no physical detail)
The rule: always give the image model something it can draw. Metaphors are great for flavor, but pair them with concrete details.
Example: Before & After
❌ Sparse (inconsistent images)
"A cool-looking guy with dark vibes"
Images will likely look inconsistent — the AI has to fill in almost everything on its own.
✅ Detailed (consistent, accurate)
"Male, early 20s. Jet-black hair, medium length, slightly messy and swept to one side. Pale skin with a sharp jawline. Dark grey eyes, intense gaze. Lean build. Wearing a fitted black leather jacket over a dark grey henley, silver thumb ring on the right hand. Gothic urban aesthetic — think rainy city streets at dusk, cool blue and violet lighting."
Notice how this covers: Form (hair, skin, eyes, build), Details (clothing, accessories), Aesthetic (setting, lighting, palette).
What to Avoid
| ❌ Avoid | Why | ✅ Use Instead |
|---|---|---|
| "Attractive" / "beautiful" / "handsome" | Too vague — the AI doesn't know what you find attractive | Describe the specific features |
| Story elements or personality | Image models can't render narratives or emotions | Save personality for the Persona field |
| Only describing the mood | Images need physical, visual details | Include mood AND concrete details |
| Very long descriptions (1000+ chars) | Can confuse image models with too many competing details | Focus on the most important 200–500 chars |
Tips by Companion Type
Realistic / Photographic
Focus on details a camera would capture:
"Woman, mid-30s. Warm brown skin, natural curly hair pulled back loosely. Deep brown eyes, gentle expression. Wearing a chunky knit sweater in cream, small gold earrings. Soft natural lighting, warm tones."
Anime / Illustrated
Include style cues alongside physical details:
"Anime-style male character. Spiky white hair, bright teal eyes, angular features. Wears a long dark coat with glowing blue accents. Cyberpunk city background with neon reflections. Clean linework, vibrant colors."
Fantasy / Supernatural
Ground the fantastical elements with specific visual details:
"Elven woman with pale lavender skin and long silver-white hair that seems to float slightly. Pointed ears, glowing violet eyes. Wearing layered robes of deep midnight blue with constellation patterns embroidered in gold thread. Ethereal forest lighting, soft mist."
Non-Humanoid / Abstract
Focus on form, texture, material, and lighting:
"A floating orb of dark amber energy, roughly the size of a basketball, with slow-moving fractures of golden light running across its surface like cracks in obsidian. Hovers about three feet off the ground. Surrounded by a faint haze of warm particles. Background: a dimly lit stone chamber."
"A wolf with midnight-blue fur that has a subtle iridescent sheen, like oil on water. Bright silver eyes. Larger than a natural wolf, with a calm, deliberate posture. Misty forest at twilight, cool blue-green lighting."
Quick Reference
| Aspect | Recommendation |
|---|---|
| Min length | 20 characters |
| Sweet spot | 200–500 characters |
| Must include | Core visual form, distinguishing details, aesthetic |
| Great to add | Lighting, color palette, textures, setting |
| Avoid | Personality traits, story elements, vague adjectives |
| Poetic language | Welcome — but always pair with something concrete |
| Experiment | Try small changes and see how they affect the images |