What to Type for AI

How to Generate Consistent Character Poses That Look Like the Same Person

One of the hardest parts of building a game character sheet with AI art is making sure your idle pose, attack pose, and hurt pose look like they belong to the same person. The character's face, proportions, and expression drift between generations, breaking immersion and forcing you to either hire a concept artist or spend hours manually editing variations.

The solution is combining two techniques: fixed seed numbers (which lock the base character's structure) and character voice anchors (which lock personality traits across every pose variation you generate). This gets you four usable poses of the same recognizable character in under an hour, instead of a weekend of trial and error.

Here's exactly how to do it.

Why Standard AI Art Prompts Fail for Game Character Sheets

When you write a new prompt for each pose, the AI treats each generation as a fresh character. Same archetype, same style keywords—but the face is subtly different. Eyes wider. Jaw softer. Hair color shifted. In a sprite sheet or character roster, players notice immediately. Your merchant and your hero suddenly look unrelated, and your four attack frames feel disjointed. Midjourney's `--seed` parameter and SDXL's seed control exist to solve this, but most tutorials skip the critical step: anchoring the character's personality and appearance *before* you vary the pose. Without that anchor, the seed alone isn't enough.

The Two-Layer Method: Seed + Voice Anchor

A character voice anchor is a short, consistent descriptor block that defines personality and look—separate from pose. Think: 'stern half-orc with ritual scars, tired but fierce, weathered hands.' You keep that block identical across all pose variations, then swap only the action description (idle, mid-sword-swing, falling backward, hand raised to speak). Paired with a fixed seed number, this locks the character's core appearance while letting you safely change body position. The result: four poses that unmistakably depict the same person in different states. No hiring an artist. No manual Photoshop work.

Step 1: Write Your Voice Anchor (Once)

Before you touch a seed, write 1–2 sentences that capture the character's essential look and energy. Include: physical marker (race, age, build), a personality detail (grim, playful, calculating), and one memorable feature (scars, distinctive armor, unusual eyes). Examples: 'young paladin with gold-traced armor and an open, determined expression' or 'elderly witch with silver braids and knowing, amused eyes.' This becomes your constant. Copy it into every pose prompt you generate.

Step 2: Generate Your First Pose with a New Seed

Write a full prompt: your voice anchor + the first pose (idle stance, calm expression, hands at rest). Generate at least 4 variations. Pick the one that feels most like your character. Note the seed number it used. You now have a locked appearance. Everything that follows uses that seed number.

Step 3: Reuse the Seed, Swap the Pose

For your second pose (attack, hurt, dialogue), write: voice anchor + new pose description + `--seed [your number]`. The AI will respect the seed and only shift the character's body position and expression relative to the action. The face, proportions, and core identity stay consistent. Repeat for every pose you need. Same seed. Different action.

Common Pitfalls and Fixes

If your second pose suddenly looks like a different person: your pose description was too vague or conflicted with the voice anchor (e.g., 'confident, heroic' + 'cowering, afraid'). Be specific about body position, not just emotion. If the character looks dramatically older or younger: you swapped the seed accidentally, or the prompt is too long (trim non-essential adjectives). If the style drifts (lighting, art quality): re-paste your style-lock keywords (e.g., 'oil painting, dramatic lighting, high detail') identically in each prompt.

Scaling to a Full Roster

One character's four poses takes roughly 10–15 minutes. A complete RPG roster of 11 characters with multiple poses and archetypes (hero, villain, NPC, enemy) takes a weekend without burning out. Each character gets its own voice anchor and seed; you organize them in a spreadsheet so you can re-roll or create variations months later without starting from scratch. This is the only way solo developers and small teams stay consistent across a full cast without outsourcing art or manual editing.

FAQ

Do I need to use the exact same prompt every time, or just the same seed?

Same seed, different pose description. Your voice anchor and style keywords stay constant; only the action changes. If you rewrite the entire prompt, you lose consistency.

What if I don't like how the character looks in the first pose?

Generate 4 variations of the first pose before you lock a seed. Pick the best one, note its seed, and use that seed for all follow-up poses. You're choosing the best version of the character, not the first one.

Can I use one seed for multiple different characters?

No. Each character needs its own seed. Reusing a seed will shift the character's face and proportions toward the original generation, even with a different voice anchor.

How specific should my voice anchor be?

2–3 defining traits: appearance, age/race, and one memorable detail. Too vague and the character drifts. Too detailed and you lose flexibility. 'Scarred veteran warrior with silver beard and tired eyes' is ideal. 'Tall, strong, brave, serious, muscular, determined, powerful' is too generic.

Does this work the same way in SDXL and Midjourney?

Yes, both support seed control. Syntax differs slightly—Midjourney uses `--seed [number]`, SDXL uses seed sliders or command syntax depending on your interface—but the principle is identical: lock the seed, vary the prompt descriptor.

Can I change the art style and keep the same character?

Yes, but carefully. If you're shifting from 'oil painting' to 'pixel art,' re-test the seed with the new style keywords. Some seeds work better in certain styles. Have fallback seeds ready.

Related

AI Product Photo Editing Prompts →Prompt guides for product photo cleanup, batch editing, consistent catalog images, mockups, and studio-ready ecommerce visuals. Stop Paying Photographers: Generate Professional Product Mockups with AI →Create 5 on-brand lifestyle mockup images per product using Midjourney or DALL-E. No studio, no photographer, no retakes. 15 ready-to-use prompts inside. How to Make AI-Generated Product Photos Look Like Real Camera Shots →Learn to generate AI product images with real camera settings so they blend seamlessly with actual photos in your shop feed. Why Your AI Product Photos Look Broken (And How to Fix Them) →Learn why AI generates melted jewelry, hallucinated text, and mismatched patterns—and the exact prompt tweaks that fix each failure. Social Content AI Prompt Growth Bundle →Social Content AI Prompt Growth Bundle combines the strongest related prompt packs into one higher-value purchase....