A game character prompt is different from a generic illustration prompt: you are not after one pretty picture, you are after a concept asset that drops into a pipeline and stays consistent across a sheet, a splash, a card, and a skin variant. The prompts that work specify three things most people skip: the game category (Western RPG / JRPG / MOBA / mobile gacha / pixel indie), the deliverable type (concept sheet, splash art, portrait card, turnaround), and the rendering style (cel-shaded 3D, semi-realistic painterly, pixel art). Ten copy-ready templates follow, plus the current consistency settings that actually hold a character together.
TL;DR
- Always name the game category + deliverable + style in the first line. Skipping any one of these is why you get generic stock fantasy.
- For character consistency in 2026, Midjourney’s old
--crefis gone in V8.1. Use Omni Reference--oref [image URL]with--ow(0–1000, default 100). Google’s Nano Banana Pro holds up to 5 characters’ identity natively. See the table in Consistency settings. - Lock one palette (2–3 colors) and one face/outfit block per character, then only vary pose, camera, and background.
- Aspect ratios:
2:3for sheets and splash,9:16for gacha cards,16:9for environment splash,1:1for pixel sprites and icons.
What separates good game character prompts
Most prompts say “a warrior with a sword.” That gives the model nothing to anchor to, so it averages every warrior in its training set. A pipeline-ready prompt covers six layers:
- Game category: Western RPG, JRPG, MOBA, mobile gacha, pixel indie, fighting game
- Class / archetype: knight, mage, archer, support, tank, assassin
- Visual style: cel-shaded 3D, semi-realistic painterly, pixel, anime key visual
- Deliverable: splash art, character sheet, portrait card, concept turnaround
- Distinctive features: silhouette readability, color accent, signature weapon / accessory
- Background treatment: gradient, environment, plain neutral
The single highest-leverage habit: write the distinctive features so the character is readable as a silhouette. Game characters live in thumbnails and busy battle scenes, so a one-line read — “platinum braid, ice-blue armor, crystal staff” — beats a paragraph of texture detail the engine will never show.
10 copy-ready prompt templates
1. Western RPG warrior (concept art)
Game character concept art of a 30-year-old female warrior, lean muscular build, short cropped dark red hair, intense green eyes, faint scar across cheek, wearing battered plate armor with leather underlayers, two-handed greatsword strapped to back, semi-realistic painterly style, plain neutral grey background, full body three-quarter view, --ar 2:32. JRPG hero splash art
JRPG-style splash art of a young swordsman with spiky silver hair and blue eyes, ornate blue and gold uniform with cape, signature crystal sword, dramatic action pose mid-leap, vibrant anime key visual rendering, fantasy battlefield background with dust and sparks, --ar 16:93. MOBA hero splash
MOBA hero splash art of a frost mage queen, regal cool expression, long platinum white braid, blue-and-silver crystalline armor with cape, summoning ice shards around her, dynamic three-quarter angle, semi-realistic painterly style with strong rim light, frozen palace background, dramatic depth, --ar 16:94. Mobile gacha portrait card
Mobile gacha character card art of a 18-year-old fox-eared girl, twin-tail platinum hair, ornate pink-and-gold shrine maiden outfit with flowing ribbons, mischievous smile, holding a glowing paper charm, soft cherry blossom backdrop with bokeh, vertical card composition with negative space at top for UI, vibrant anime style, --ar 9:165. Pixel art hero (indie game)
16-bit pixel art character sprite of a wandering monk, brown robes, straw hat, wooden staff, walking cycle frames on a transparent background, clean limited 16-color palette, top-down adventure RPG style, --ar 1:16. Fighting game roster portrait
Fighting game character roster portrait of a 25-year-old martial artist, athletic build, short black hair, fingerless gloves, tight black training pants and red sash, fierce ready stance, full body, vibrant cel-shaded 3D anime style, plain gradient backdrop, --ar 2:37. Tank / heavy class concept
Concept art of a heavy tank class character, hulking armored figure with massive shoulder pauldrons and tower shield, helmet with single eye slit, dark steel and crimson color scheme, low angle imposing pose, gritty Western RPG painterly style, plain dark grey background, --ar 2:38. Stealth / assassin character
Concept art of a stealth assassin, lean female figure in dark grey leather armor with hood, twin curved daggers, half-shadowed face revealing only sharp eyes, crouched ready pose on a rooftop edge, moonlit cityscape blurred in background, semi-realistic painterly style, --ar 2:39. Mage / spellcaster
Concept art of an elemental fire mage, 28-year-old male, long auburn hair, intricate red and gold robes with rune embroidery, casting a flame spiral around hands, glowing eye effect, semi-realistic painterly style, dramatic side lighting, plain dark background, --ar 2:310. Stylized chibi gacha collection card
Chibi-style gacha collection card art, big-head small-body knight character with oversized helmet and tiny sword, cheerful pose, bright colorful background with sparkles, vector-clean rendering, mobile game card composition, --ar 9:16Per-deliverable tuning
- Concept sheet (turnaround):
front, side, back views on a single canvas, T-pose, plain neutral grey background, no shading variation - Splash art: action pose, dramatic lighting, complex background
- Portrait card: vertical, negative space at top for UI, soft background bokeh
- Pixel sprite: limited palette count, transparent background, walking-cycle frames
- Hero turnaround for 3D handoff: orthographic view, neutral lighting, no rim
Which tool for which job
You do not need the most expensive plan to start. As of June 2026:
| Tool | Entry price | Best at | Commercial use |
|---|---|---|---|
| Midjourney V8.1 | Basic $10/mo ($8 annual) | Splash art, painterly anime/RPG, Omni Reference consistency | Commercial rights on all tiers; Pro/Mega required above $1M revenue |
| Nano Banana Pro (Gemini 3 Pro Image) | In Google AI Pro $19.99/mo | Multi-character scenes, 4K, locked identity for up to 5 characters | Allowed under Google’s terms; verify per use |
| Stable Diffusion / FLUX.2 (local or hosted) | Free locally; hosted varies | LoRA-trained heroes, pixel/sprite work, full control | License depends on the specific checkpoint |
Midjourney’s Standard plan ($30/mo) unlocks unlimited Relax-mode generations, which matters once you are iterating dozens of variants per character. See Midjourney’s plan comparison and Google’s Nano Banana Pro page for current details. Whatever the tool, AI is best for fast concept exploration; most studios still have an artist repaint the final sheet to dodge copyright risk before it enters production.
Common mistakes
- No game category: you get generic averaged fantasy
- Cluttered weapon / accessory description: the model can’t fit them all readably
- Multiple style words conflicting (
pixel art + cel-shaded + photorealistic): the model picks one at random - Forgetting deliverable type: splash vs. portrait vs. card each need different composition
- Asking for “epic” or “legendary”: empty words with no visual anchor
Consistency settings (June 2026)
Holding one character together across a sheet, splash, card, and skin variant is the hard part of game art. The fastest method changed in 2026, so the old advice (“use --cref”) is now wrong on the current Midjourney model.
- Lock a face/outfit block. Write the hair, eyes, face, signature outfit, and palette as one fixed paragraph and paste it verbatim into every prompt. Only vary pose, camera, and background. This alone gets you ~80% of the way.
- Midjourney V8.1 uses Omni Reference, not
--cref. As of June 2026 Midjourney ships V8.1 (released April 30, 2026), and the classic--crefis incompatible with V7 and later. Append--oref [image URL]with one reference image, then tune--ow(omni-weight, 0–1000, default 100): lower values let the style shift, higher values lock the character’s face and outfit. Omni Reference costs ~2x GPU time and does not run in Fast/Draft mode. - Reuse a seed. V8.1 brought back a working
--seed; fix it so re-rolls of the same prompt land near the same composition. - Nano Banana Pro for multi-character scenes. Google’s Nano Banana Pro (the Gemini 3 Pro Image model) holds the identity of up to 5 characters and the fidelity of up to 10 objects in one workflow at up to 4K, which makes it strong for party shots and storyboards where Midjourney drifts.
- Train a LoRA for hero characters. For a protagonist that appears in hundreds of assets, a small LoRA on 20–30 turnaround images beats prompt tricks for true consistency.
See How to Keep AI Image Style Consistent for the full reference-image and seed workflow.
Consistency methods at a glance
| Method | Best for | Effort | Consistency |
|---|---|---|---|
| Fixed face/outfit block | Every project, first pass | Low | Medium |
Midjourney --oref + --ow | One hero across poses | Low | High |
| Nano Banana Pro multi-ref | 2–5 characters in one scene | Medium | High |
| Trained LoRA (20–30 images) | Protagonist in 100+ assets | High | Very high |
FAQ
Q: Can I use AI character concepts in a commercial game? A: Check your tool’s license. Midjourney grants commercial rights on every paid tier as of June 2026 (companies over $1M revenue must be on Pro or Mega). Locally run Stable Diffusion / FLUX depends on the specific checkpoint license. Most studios still have an artist repaint AI concepts before shipping to reduce copyright risk, and you should do a legal review for any character that goes into a released game.
Q: How do I get a consistent character across different poses?
A: On Midjourney V8.1 use Omni Reference (--oref [image URL] plus --ow 0–1000) — the old --cref no longer works on V7 and later. Nano Banana Pro keeps up to 5 characters’ identity natively. For a hero in hundreds of assets, train a LoRA. A single prompt with only seed-lock is not enough across very different poses.
Q: Best aspect ratio for game character art?
A: 2:3 for character sheets and splash, 9:16 for gacha cards, 16:9 for environment splash, 1:1 for pixel sprites and UI icons.
Q: How do I get pixel art that’s actually pixel-perfect? A: Generate at small resolution, then downscale and palette-reduce in a dedicated tool like Aseprite. Diffusion models output “pixel-style” images with anti-aliasing and off-grid pixels, so a post-processing pass to a fixed grid and limited palette is what makes them genuinely pixel-perfect.
Q: Why does my character have 7 fingers or two left hands? A: Hands are the perennial AI weakness, though Midjourney V8.1’s anatomy is noticeably better than V6. Either redraw the hand in post, or compose poses where hands are partially hidden, behind the back, or holding a weapon that occupies the fingers.
Related articles
- Best Anime Character Image Prompts
- Realistic Handsome Man Portrait Prompts
- Realistic Beautiful Woman Portrait Prompts
- How to Keep AI Image Style Consistent
- How to Fix Distorted Faces in AI Images
- Fantasy Character Prompts: 12 TTRPG & Game Splash Templates
- Game Character Class Prompts: Warrior, Mage, Rogue Archetypes