Model Guide

How to Use GPT Image 2 for Princess Mononoke Key Visuals

A three-lesson workflow for turning the world's cards into key visuals: San on the root, the Sacred Pool at night, and Iron Town at dusk.

How to Use GPT Image 2 for Princess Mononoke Key Visuals

This guide teaches the model through the world. Instead of a list of magic phrases, you get three lessons — a character, a location and a town — each built from a card, a beat and a camera. By the end you should be able to write a Mononoke key visual prompt in about four minutes and repair it in one pass.

It assumes no prior knowledge of the film and no prior experience with image models. You need the world's cards, GPT Image 2, and one decision per lesson: what is happening in the next ten seconds.

GPT Image 2

San, the wolf-raised girl, standing on a moss-covered cedar root at dusk with a spear, red war paint on her cheeks and two white wolves behind her

GPT Image 2, San character key visual: the pelt reads as weight, the spear as a tool, and the dusk sky gives the ravine depth behind her.

GPT Image 2

The Great Forest Spirit, a towering deer-like god with a human face and branching antlers, standing in a moonlit pool surrounded by small kodama spirits

GPT Image 2, Sacred Pool location key visual: the god is small enough to be in the water and large enough to fill the canopy, with kodama on every rock.

What You Need Before You Start

The world cards

Eight characters, five locations, five events, four factions and four rules. You do not need all of them. You need one character or place, plus one rule that constrains it.

GPT Image 2 access

Text input only is fine. Reference images are supported but not required for any lesson here.

Two frame sizes

1024×1536 for characters and posters, 1536×1024 for places and crowds. Decide before you write the prompt, because the crop is part of the composition.

One exclusions line

"No text, no watermark, no modern objects." Paste it every time. It prevents most of the boring failures.

Understand the World's Visual Grammar

The four rules are also four image instructions. Learn them once and you stop needing to describe the setting from scratch.

The Demon Curse → a wound that moves

Black tendrils under the skin, a scar that climbs the arm, strength out of proportion to the body. If a frame includes Ashitaka, the arm is a compositional element, not a detail.

Iron Against Forest → two materials in one frame

Smoke against canopy, cut wood against moss, iron against hide. Every strong image in this world contains both materials and lets them touch.

The Law of the Forest Spirit → two silhouettes, one being

A deer with a human face by day; a translucent colossus by night. Choose which version the scene needs before writing a single adjective.

A Head for Immortality → hunters who are afraid

Rope, nets, guns, a head treated as cargo. The rule turns belief into visible equipment. Use it whenever you need human antagonists without making them cartoon villains.

The grammar also has a colour rule: earth palette for the human world, one cold emission for the sacred. Moss, clay, iron, undyed cloth — then a pale blue-white that does not obey the sun. If your image has no impossible light, it is a historical scene, not a myth.

Build the Prompt in Five Layers

Five layers, each doing a different job. If you only remember one thing from this guide, remember the order.

Layer 1: style and medium

Hand-painted anime cel, flat colour fields, visible brush edges, muted earth palette. This must come first or the detail clauses will override it.

Layer 2: subject and costume

Who, wearing what, holding what. Use the card's exact nouns: wolf-pelt hood, red war paint, bone dagger, spear.

Layer 3: companions and crowd

Two wolves behind her, kodama on the rocks, women at the bellows. Companions establish scale and social context.

Layer 4: place, weather, light

One place, one time of day, one named light source. Add the impossible light here, not at the end.

Layer 5: camera and exclusions

Low or high camera, vertical or wide, no text, no watermark, no modern objects. This is the layer that fixes most bad crops.

[style] Princess Mononoke [shot type] key visual, hand-painted anime cel, muted earth palette, [aspect]: [subject + costume + props]; [companions]; [place + weather]; [named light source + impossible light]; [camera height + composition]; no text, no watermark, no modern objects.

Lesson 1: San on the Root

Read the card

Pelt hood and cloak, red paint on both cheeks, bone dagger, spear, raised by wolves. The card's hidden instruction is hierarchy: the wolves are larger than she is.

Choose the beat

Dusk, before a raid, at the edge of the forest. Standing, not posing. Looking out of frame, not at the camera.

Write five layers

Cel style; San in pelt with spear and dagger; two white wolves behind her in mist; moss-covered root above a ravine; warm western horizon plus cold green shadow; low camera, vertical.

Generate at 1024×1536

High quality. Check hands, paint, and whether the wolves are behind her. A wolf beside her changes the story from pack to pet.

Repair one clause

If the pelt reads as a cloak, replace "pelt hood" with "animal hide with visible fur break and ragged hem". If the spear is vague, add "wrapped grip, leaf-shaped iron head".

The GPT Image 2 result is a three-plane scene: root, girl, ravine. The Krea 2 result on the same prompt is a flatter night poster with a red spear shaft and glowing mushrooms. Both are valid. The GPT version explains the place; the Krea version sells the character at a glance.

Lesson 2: The Sacred Pool

Pick the rule and the hour

The Law of the Forest Spirit. Night gives you the walker and the emission; day gives you the deer and the human face. This lesson uses night with the human face kept.

Set the camera at water level

Low camera, wide frame. The pool becomes a mirror and the god becomes both body and reflection.

Name two lights

Moonlight through cedar and the god's own pale blue emission. Two sources, one cold, one colder. Without both you get blue fog.

Add small lives for scale

Kodama on the mossy rocks, some near the camera, some far. They give the god an audience and a size.

Generate at 1536×1024

Check the face first, then the antlers, then the waterline. If the face has become a deer, repeat "calm human face with dark eyes" and reduce the antler count.

GPT Image 2 fills the canopy with antlers and scatters kodama across the whole frame. Krea 2 comes closer, simplifies the crown and lines the kodama up like a chorus. For a location that must feel vast, take the GPT frame. For a cover that must read at 300 pixels, take the Krea frame.

Lesson 3: Iron Town at Dusk

Choose the social fact

Iron Town is a refuge funded by deforestation. The image must show both: warm human industry and the cut forest behind it.

Build the layers of the town

Wall, gate, wooden scaffolding, furnace chimney, women at the bellows, rifle racks, lepers on the upper terrace. Pick three, not seven.

Put one figure above the town

Eboshi on a walkway with the furnace behind her. Dark kimono, short rifle, soot on the forearms. She should be the only still thing in the frame.

Light it from inside

Furnace orange as the key, cold blue dusk as the fill, black smoke against a green-black ridge. The town should be the warmest object in a cold world.

Generate wide, crop tight

Start at 1536×1024 to establish the geography, then crop a portrait for the cover. Do not generate the portrait first and try to add a town behind it.

This is the lesson where GPT Image 2's depth matters most. A forge town is a stack of planes — near scaffolding, middle gate, far furnace, forest behind — and a flat composition turns it into a wall of brown. If the first result is a wall, add "foreground walkway in shadow, middle ground gate lit by furnace, background ridge and smoke" and run it again.

Iterating Without Losing the World

Change one layer per run

Change the beat and keep the light, or the light and keep the beat. Two changes at once and you cannot tell which one broke the image.

Keep a fixed palette line

Paste the same palette clause into every prompt in a sequence: moss green, iron black, clay red, undyed cloth, one pale blue emission. Consistency is a text artifact before it is a visual one.

Keep the camera rule

If the sequence is about human smallness, every frame stays low and wide. A single eye-level close-up will read as a different world.

Save the clause that worked

When a phrase fixes a problem, add it to the card as a visual note. The next prompt starts from a better place.

Reading a Failed Generation

It looks like a photograph

Style arrived too late or detail words arrived too early. Move the cel clause to the front, cut texture adjectives, add "flat colour fields, visible brush edges".

It looks like a different anime

The palette clause is missing. Add earth tones and one cold emission; remove saturated blues, pinks and rim-light language.

It looks like a park

Scale is wrong. Add a measurable comparison and push the canopy out of frame. A person at one-third height fixes most forest images.

It looks like a costume shop

Too many props. Keep the two that carry the beat and delete the rest. A spear and a dagger are enough for San; a necklace, mask and bow are not.

GPT Image 2 vs Krea 2: Which for Which Shot

Character sheetGPT Image 2 for the detailed version, Krea 2 for the graphic version. Run both from the same prompt; use the GPT one to check costume logic, the Krea one for the cover.
Location establishing shotGPT Image 2. It builds depth, a far bank, a waterfall, a canopy. Krea 2 flattens the same scene into bands and loses the geography.
Social media cropKrea 2. Strong centred silhouette, two or three values, readable at small size, and about five times cheaper per image.
Anything with a face that must be wrongGPT Image 2. It will hold the human face on a deer, the fear on a hunter, the soot on Eboshi. Krea 2 tends to prettify and soften.

Publishing and Reusing

Publish the image with the card it came from, not as a standalone picture. A viewer who can open San's card, read the pelt and the paint, and then see the image will understand why it looks the way it does. That link between text and image is the difference between a world and a gallery. Keep the world's independent, non-affiliated note with any reuse, and do not present generated frames as official stills.

Checklist Before You Publish

One beat

Can you say what is happening in the next ten seconds? If not, the image is a portrait, not a scene.

Two materials

Does the frame contain both the human world and the forest world — smoke and moss, iron and hide, cloth and bark?

One impossible light

Is there a cold emission that does not obey the sun? If everything is lit naturally, the myth is missing.

Correct scale

Is the human smaller than the forest? Are the wolves larger than the girl? Does the god fill the canopy?

Two props maximum

Can you name the props without counting on one hand? If not, remove until you can.

No modern objects

Check zips, modern firearms, plastic, electric light and text. One anachronism breaks the frame faster than a bad hand.

Questions About This Workflow

How long should one image take?

About twenty minutes once you know the world: three minutes choosing the argument, four writing the layers, five generating on both models, five repairing one clause, three filing the result back into the card. The first image takes longer because you are learning the palette.

Do I need to prompt in English?

Both models were tested in English and the clause order assumes English weighting. If you write in another language, keep the style clause first and the exclusions last; the middle layers are more forgiving than the ends.

Use the guide on the real world

The cards, the images and the comparison are all in one place. Generate your first key visual — or open the studio and pick a scene the guide does not cover.

Create From This World Explore the World Studio

These key visuals were generated from this world's materials with GPT Image 2 and Krea 2. Princess Mononoke is an independent, fan-built world in Neta Studio and is not affiliated with Studio Ghibli. The key visuals on this page were generated inside this world; every character, place and event is an editable material rather than canonical text. Open the studio to read every card.