A class portrait succeeds when the silhouette tells you the class before the colors do: a warrior reads as a triangle, a mage as a column, a rogue as a coil. AI default poses are static and frontal, which flattens all three into the same rectangle. The 12 prompts below force a three-quarter angle, distinct silhouette cues, class-specific lighting, and a single signature element per class, so warrior, mage, rogue and the rest stay visually separable in a party shot. For full game key art, pair with game character design prompts.
TL;DR: Copy the prompt for each class, generate in Midjourney V7, Nano Banana 2 (Gemini), or GPT Image 2, then lock one camera angle and lighting direction across all twelve so the party reads as one game world. For repeatable identity across renders, use Midjourney’s Omni Reference (--oref + --ow, V7 only) or upload a fixed reference in the Gemini app. Pricing and model versions below are current as of June 2026.
Best for
- Game character art and TTRPG party portraits
- Indie game key art and Steam capsules
- Card game illustrations where the class must read at thumbnail size
- Worldbuilding moodboards for an upcoming campaign
- VTuber and stream avatar variants tied to a class fantasy
Which model to use (June 2026)
All twelve prompts are written as plain natural-language descriptions, so they run on any current image model. The differences that matter for a class roster are art style and how the tool keeps a face consistent across a party.
| Tool | Best for | Entry price | Identity control |
|---|---|---|---|
| Midjourney V7 | Painted concept-art look, strongest stylized output | $10/mo Basic (~200 images); annual ~$8/mo | Omni Reference --oref + --ow (1-1000) |
| Nano Banana 2 (Gemini 3.1 Flash Image) | Fast iteration, multi-character party shots | Google AI Pro $19.99/mo; limited free tier | Up to 5 consistent characters in the Gemini app |
| GPT Image 2 (ChatGPT) | Easiest editing in plain chat, conversational fixes | Bundled with ChatGPT Plus $20/mo | Reference upload + iterative edits |
| Flux 2 (API/hosts) | Photoreal renders, lowest per-image cost | ~$0.01-0.10 per image, pay-as-you-go | Reference image conditioning |
For stylized fantasy class art, Midjourney V7’s painterly aesthetic is still the leader as of June 2026. If you want to iterate fast or keep several party members consistent in one prompt, Nano Banana 2 became the default Gemini image model in February 2026 and holds up to five characters in a single generation. Note that Midjourney’s old --cref flag does not work in V7; Omni Reference (--oref) replaced it.
1. Warrior
fantasy warrior character, heavy plate armor with weathered scratches and dents, two-handed sword strapped across the back, scar across the cheek, dramatic three-quarter angle, low camera, painted concept art style, triangular silhouette2. Mage
fantasy elemental mage, flowing dark robe with arcane silver embroidery, glowing staff held vertical, faint runes orbiting the open hand, mystical rim lighting from behind, painterly concept art, columnar silhouette3. Rogue
fantasy rogue, dark leather hood half-shadowing the face, twin daggers reversed at the hips, sly half-grin, dramatic torch backlight from a narrow alley, asymmetric coiled silhouette, painted concept art style4. Ranger
fantasy ranger, hooded green cloak, longbow in one hand and quiver across the back, weathered boots, forest canopy filtering dappled light from above, three-quarter angle, painterly concept art, lean elongated silhouette5. Paladin
fantasy paladin, ornate silver plate armor with engraved sigils, two-handed warhammer planted point-down at the feet, soft holy light from above-camera, three-quarter angle, stoic expression, painted concept art, broad and grounded silhouette6. Druid
fantasy druid, layered earth-tone robes with living vines woven through, wooden staff topped with a small living plant, antler circlet, soft forest light, woodland creature partially visible at the feet, painterly concept art7. Monk
fantasy martial monk, simple wrapped robe with exposed forearms, hands raised in a relaxed fighting stance, no weapons, mountain temple courtyard in soft focus behind, low sun rim-light, painted concept art, balanced compact silhouette8. Necromancer
fantasy necromancer, dark hooded robe with bone clasps, skeletal staff with a faintly glowing green orb, two small wraith wisps orbiting the shoulder, low-key crypt lighting from below, painterly concept art, angular silhouette9. Bard
fantasy bard, colorful traveling clothes with mismatched layers, lute slung across the back, mid-flourish gesture as if mid-song, warm tavern firelight behind, three-quarter angle, painted concept art, dynamic open silhouette10. Berserker
fantasy berserker, fur and leather harness over bare scarred torso, twin axes held wide, war-paint across the eyes, snowstorm whipping the hair sideways, low angle, painted concept art, aggressive widened silhouette11. Alchemist
fantasy alchemist, long leather apron over travel clothes, glowing vial in one hand and a closed grimoire under the other arm, brass-and-glass gadgets at the belt, candlelit workshop softly out of focus, painterly concept art, cluttered functional silhouette12. Summoner
fantasy summoner, ceremonial robe with summoning glyph patterns, small spirit familiar materializing at the shoulder, hand outstretched with glowing sigil hovering in the palm, twilight sky behind, painterly concept art, asymmetric silhouetteHow to refine
Lock the camera angle and the lighting direction across all twelve renders, then iterate only on the subject. That is how a party shot feels like one game world instead of twelve different artists. If you are building a campaign or capsule set, render each character with the same style reference so the visual system stays consistent.
In Midjourney V7, add --sref <image_url> to pin one shared art style across the whole roster, and set a fixed --seed so re-rolls stay stable. To keep a single character’s face and outfit consistent across multiple poses, use Omni Reference: --oref <image_url> --ow 200. Push --ow toward 400 when the face or costume must be preserved exactly, and keep it below 400 otherwise, because very high weights make results unpredictable. In the Gemini app, upload the same reference image and Nano Banana 2 will hold up to five characters consistent in one shot. The AI consistent character images tutorial covers the full seed-and-reference workflow for series consistency.
Common mistakes
- No armor wear or damage texture — characters look like cosplayers, not adventurers
- Static frontal pose — the silhouette collapses to a rectangle and class no longer reads
- Generic forest backgrounds for every class — a necromancer in a sunlit forest reads wrong
- Mixing painted concept art with photo-real rendering across the party — pick one style and lock it
- Same lighting for every class — paladin and necromancer should not share the same light recipe
FAQ
Which model gives the best fantasy class art in 2026? For a painted concept-art look, Midjourney V7 is still the strongest as of June 2026, starting at $10/month on Basic. For fast iteration or multi-character party shots, Nano Banana 2 (Gemini 3.1 Flash Image) is the default Gemini image model and holds up to five consistent characters per generation. GPT Image 2 inside ChatGPT Plus ($20/month) is the easiest if you want to edit a portrait conversationally.
How do I keep the same character across multiple poses?
In Midjourney V7, use Omni Reference: --oref <image_url> --ow 200, raising --ow toward 400 when the face and outfit must match exactly. The older --cref flag does not work in V7. In the Gemini app, upload one reference image; Nano Banana 2 keeps up to five characters consistent in a single prompt.
Do these prompts work outside Midjourney?
Yes. Every prompt is plain natural language with no Midjourney-only syntax, so it runs as-is in Nano Banana 2, GPT Image 2, Flux 2, and Stable Diffusion front-ends. Only the consistency flags (--oref, --sref, --seed) are Midjourney-specific.
Why force a three-quarter angle instead of a front view? A frontal pose collapses every class into the same rectangle, so the silhouette stops carrying information. A three-quarter angle shows the two-hander on a warrior’s back, the coiled crouch of a rogue, and the column of a mage’s robe, which is what makes the class readable at thumbnail size.
How do I keep a 12-character party looking like one set?
Lock one camera angle, one lighting direction, and one art style across all twelve renders, then change only the subject. In Midjourney, pass the same --sref style reference and a fixed --seed to every prompt so the whole roster reads as one game world.
Related
- Game character design prompts — broader character-design library
- Game character portrait sheets — turnaround / sheet patterns
- Fantasy character prompts — fantasy genre library
- AI consistent character images — tutorial for series consistency
- Mobile Gacha Character Prompts: 10 SSR Banner Templates
- RPG Hero Class Prompts: 10 Party Role Templates
Tags: #Game character #Fantasy