GPT Image 2 Zelda World Generator — Key Visuals From a Kingdom, Not a Keyword
A working GPT Image 2 generator for Zelda-style worlds: the three-pass prompt, the Hyrule style contract, the literal prompts, and an honest comparison with Krea 2 on the same two briefs.
GPT Image 2 is a general image model that becomes a Zelda world generator when you give it a world instead of a mood. Asked for 'Zelda art', it gives you a green hat and a sword in a field. Given a specific kingdom, a failing seal, a plateau, a cocooned castle, four machine silhouettes and a style contract, it produces a frame you could use as a world key visual — the kind of image that makes a campaign feel like it already exists.
Everything below was generated inside Hyrule: The Blood Moon Age on Neta Studio: four published works, two prompts, two models (GPT Image 2 and Krea 2), same 1024×1024 size, no post-processing. The world supplies the canon and the style contract; the model supplies the decisions. This page is the working manual for that division of labour, with the exact prompts published in full.
| Quick facts | |
|---|---|
| Model | GPT Image 2 |
| Capability | Zelda / dark fantasy world key visuals and character posters |
| World | Hyrule: The Blood Moon Age (26 atoms) |
| Prompts | 2 — a world establishing frame and a character poster |
| Settings | 1024×1024, quality high, no reference images |
| Comparison | Krea 2 on the identical prompts, same size |
| Published works | 4 — two models × two prompts |
| Best at | Depth, weather, materials, multi-source light, scene truth |
What GPT Image 2 Does With a Kingdom
The model has three habits that matter for world art. First, it treats a prompt as a scene brief: name a plateau, a cocooned castle, four machine silhouettes and dead Guardians in the grass, and it will render all of them in a believable depth order. Second, it resolves materials: stone reads as stone, cloth as cloth, metal as metal, and it will happily give you six kinds of shadow in one frame. Third, it composes rather than decorates — it decides where the eye enters, what the middle distance holds and what the background owes the foreground.
Those habits are why it suits a world page. A world key visual has to do a lot of work in one image: establish geography, state the conflict, suggest a history and leave room for a story. GPT Image 2 is unusually willing to carry all four. The cost is speed and a tendency to over-detail; the fix is a narrower prompt and a style contract that forbids micro-texture.
The Three-Pass Prompt
Write the prompt in three passes, in this order. Pass one is the world: where you are, what the weather is, what the light source is. Pass two is the story: what is wrong, what has already happened, what the viewer should worry about. Pass three is the style contract: medium, edges, colour, detail budget and the negative list. If you write the style first, the model gives you a beautiful empty frame. If you write the story first, it gives you a scene with no world.
Geography
Name the viewpoint and the depth stack: foreground, middle distance, horizon. Say what is in each band.
Weather and light
One moon, one volcano glow, one shrine cyan. Give the model a single direction and let it place the rest.
History
Add the evidence: dead Guardians, a broken bridge, a road gone to grass. History is what the frame has already survived.
Style contract
Medium, shaping, edges, colour groups, detail budget, negative list. Keep it under 80 words.
One frame only
No collage, no split screen, no 'concept sheet'. One camera, one moment.
The Style Contract for Hyrule
This world uses a style called Commercial Game Painterly. It is narrow on purpose. Broad visible painted planes, hard brush-shaped edges at structural turns and focal points, large connected colour groups, deep-colour shadows and a few saturated accents. Detail is spent on identity, hands and signature props; large surfaces stay quiet. The negative list bans line art, airbrush gradients, PBR reflections, photographic skin, 3D render look, text, logos and watermarks.
| Quick facts | |
|---|---|
| Medium | High-finish commercial game painting |
| Shaping | Broad visible painted planes, strong structural grouping |
| Edges | Hard at turns and focal points; soft only on organic and distant forms |
| Colour | Large connected local colour, deep shadows, few saturated accents |
| Detail | Face, hands, signature props; large surfaces quiet |
| Negative | Line art, gradients, PBR glass, photo skin, 3D look, text, logos |
World Key Visual — The Literal Prompt
This is the exact prompt used for both world images, at 1024×1024, with no reference image. GPT Image 2 and Krea 2 received it character for character.
Key visual for the dark fantasy game world "Hyrule: The Blood Moon Age". A single wide panoramic establishing landscape at night with no characters: a vast green kingdom of hills and dead roads under an enormous blood-red moon; in the middle distance Hyrule Castle stands wrapped in a red-black cocoon of Malice, its spires pierced by glowing veins, the ruined Castle Town spread at its feet; to the left the high cut-off table of the Great Plateau with a ruined temple and a cold blue shrine-light; to the right a volcano with a red caldera and a distant desert wall; four mechanical Divine Beasts silhouetted on the horizon like patient animals; broken Guardian machines half-buried in grass in the foreground, their orange eyes dim; thin red embers drifting in the air; mist in the valleys. High-finish commercial game painting, shaping form with broad visible painted planes and strong structural grouping, solid brush-shaped hard edges at structural turns and focal points, limited soft edges on organic and distant areas, large connected local color and value groups, deep-color shadows, a few saturated accent colors of blood red and shrine cyan, detail concentrated on identity and signature props, large surfaces quiet and connected. Cinematic lighting, clear focal subject and readable depth layers, epic and oppressive mood. No text, no letters, no logo, no watermark, no UI, no people, no creatures.
Character Key Visual — The Literal Prompt
The character prompt adds one figure and two competing lights. Everything else stays in the world contract: the plateau, the cocooned castle, the blood moon, the dead Guardians. The model is not asked to invent a Hyrule; it is asked to put one person in the Hyrule that already exists.
Character key visual for the dark fantasy game world "Hyrule: The Blood Moon Age": Link, the knight who was supposed to die a hundred years ago. A single figure, full body, standing on a windswept cliff edge of the Great Plateau at night, three-quarter view, facing the distant castle. He is a lean young man with long ash-blond hair tied back, a scarred right cheek, a worn and repaired blue Champion's tunic with faded white embroidery, leather bracers, a traveler's bow and quiver on his back, and the Master Sword held low in his right hand, its blade faintly blue and chipped; his left hand is wrapped in a fingerless glove marked with a dimly glowing Sheikah eye. Behind him the enormous blood-red moon rises over Hyrule Castle, wrapped in a red-black cocoon of Malice with glowing veins; red embers drift past his shoulder; dead grass and broken Guardian plating lie at his feet; cold cyan shrine-light rim-lights his left side, warm red moon light his right. Determined, exhausted, unsmiling expression. High-finish commercial game painting, shaping form with broad visible painted planes and strong structural grouping, solid brush-shaped hard edges at structural turns and focal points, limited soft edges on organic turns, large connected local color and value groups, deep-color shadows, a few saturated accents of blood red and shrine cyan, detail concentrated on face, hands and the sword, large surfaces quiet and connected. Cinematic character poster, clear focal subject and restrained background, vertical hero composition. No text, no letters, no logo, no watermark, no UI, no extra characters, no monsters.
GPT Image 2 vs Krea 2 on the Same Brief
This is not a winner-takes-all comparison. The two models answer different questions. GPT Image 2 answers 'what would it feel like to stand there?' Krea 2 answers 'what would this look like on a shelf?' A world pipeline needs both answers, and publishing the pair is more useful than picking a champion.
| Quality | Stronger model | What you actually see |
|---|---|---|
| Depth and distance | GPT Image 2 | Krea 2 flattens the middle distance into a graphic band |
| Materials | GPT Image 2 | Metal, cloth, wet stone and ember light stay distinct |
| Silhouette | Krea 2 | Cleaner shapes, stronger read at 200 pixels |
| Colour | Krea 2 | Flatter, brighter, more poster-like groups |
| Faces and gear | GPT Image 2 | Scars, repairs, straps and glove detail survive |
| Speed and cost | Krea 2 | Faster and cheaper per image |
| Consistency across a set | Both | Same prompt and size; the pair becomes the canon |
For this world, the practical rule is: generate the establishing shot on GPT Image 2, generate the cover on Krea 2, and publish both with the same prompt. The difference is not noise; it is art direction you can point at. When a later image drifts, compare it to the pair and you will know which axis moved.
Sizes, Quality and What to Spend
- 1024×1024 The comparison size used here. Safe for both models and honest for side-by-side publication.
- 1536×1024 GPT Image 2 only. Better for a wide hero or a landing-page banner; Krea 2 cannot match the aspect.
- quality=high Worth it for a world key visual that will be inspected. Skip it for iteration.
- quality=medium The iteration setting. Generate four, pick one, then re-run the winner at high.
- No reference images This comparison deliberately uses text only, so the difference is the model rather than the input.
- One prompt, two models The cheapest way to learn a model's taste. Two images teach more than ten prompts on one model.
Common Failure Modes, and the Fix
Most bad Zelda world images fail for one of five reasons, and none of them require a new model. They require a narrower frame. Fix the prompt before you fix the sampler.
From Image to World Page
An image is not a world. The step that turns one into the other is association: the work gets a title, a prompt, a model, a summary and a link to the world. On Neta Studio that means the image is published as a work inside Hyrule: The Blood Moon Age, with the world's 26 atoms behind it. The page you are reading is the proof that the association holds.
Keep the prompt
Store the exact text and the model name with the image. Without them the image is a screenshot, not a work.
Title it like a scene
'Hyrule Under the Blood Moon — World Key Visual (GPT Image 2)' beats 'zelda_final_v3.png'.
Write a 40-word summary
What the image shows and what the model did. This becomes the share description.
Attach it to the world
The work must belong to the world, not to a folder. The world page is the canonical index.
Publish the pair
If you ran a comparison, publish both and say which is which. The comparison is the content.
A Prompt Library for Hyrule
- World, default A wide establishing landscape at night, the cocooned castle at centre, the plateau shrine-light left, volcano right, four Divine Beast silhouettes on the horizon, dead Guardians in the foreground grass, blood moon, embers, mist in the valleys.
- World, weather variant The same frame in freezing rain: the castle lights smeared, the plateau invisible, the Guardians half-drowned, one shrine cyan cutting through.
- Character, knight One figure, full body, three-quarter view, repaired blue tunic, Sheikah glove, chipped Master Sword low, blood moon behind, shrine cyan rim light, dead grass and broken Guardian plating underfoot.
- Character, princess Zelda inside the seal: one figure, hands against a wall of red-black Malice, gold light leaking between her fingers, no crown, no throne, exhausted.
- Faction, Gerudo A mudbrick gate at dusk, cold fountains, a thunder helm on a stand, one guard looking the wrong way.
- Relic, sword The Master Sword in stone in a fog-locked forest, one shaft of light, no character, no text.
Questions About the Generator
Generate a Hyrule you can play in
Open the world, copy the style contract, and run the three-pass prompt. Start with the establishing frame, then the character, then the comparison. The prompts on this page are copy-paste ready.
GPT Image 2 + Krea 2, one world
Two models, two prompts, four published works — the exact prompts are on this page.
Platform: Neta Studio · Type: Tool · GPT Image 2 Zelda world generator · World: Hyrule: The Blood Moon Age · Works: Hyrule Under the Blood Moon (GPT Image 2 / Krea 2) · The Knight Who Should Have Died (GPT Image 2 / Krea 2)