🎭 Character Design with Nano Banana 2

Reference 1 locks in the identity: facial structure, facial proportions, skin tone, hairstyle, and the character's recognizable appearance.

It’s what ensures that, in any style, it’s still the same person. Without it, each generation becomes a different character.

Identity = face + proportions + skin + hair + “look.” Never delegate this to Reference 2.

Reference 2 guides only the outfit's design: silhouette, fabric type, accessories, color palette, and styling mood.

You borrow fashion language without importing the face from Ref 2—so the outfit changes, but the person doesn’t.

"Use Reference 2 ONLY for the outfit." That sentence appears in every prompt for a reason.

Each prompt explicitly instructs you NOT to copy the pose from any reference and to create a new, specific pose.

It’s what sets “moving the character through worlds” apart from “repainting the same image.” The new pose adds life and variety.

Describe the pose in detail (sitting, contrapposto, three-quarter view) — vague prompts produce generic results.

The deliberate separation between who the person is (Ref 1) and how the world is styled (Ref 2 + genre).

When identity and clothing inspiration get mixed together, the results become random. The split keeps everything predictable.

One prompt = two clear sources. Mix them up and you lose control.

The workflow: keep the face consistent, change the visual genre, and create a new pose every time.

It’s what makes the prompts useful for repeatable character design at scale — a brand, an avatar, a series.

Lock face → change genre → new pose. Three steps, endless variations.

The system is consistent because the face comes from one reference and the fashion from another — without locking the image into a single pose.

Understanding the “why” lets you adapt the system to any new style, not just the 10 in the pack.

Strong identity + strong visual genre = a result that’s recognizable and varied at the same time.

Surreal editorial oil painting: vintage lounge, bizarre animals, expressive brushstrokes, confident seated pose.

Shows that the system works even in sophisticated, absurd-at-the-same-time “gallery” styles.

Lowbrow + café culture + 4:5 vertical. Full prompt in the module.

Ugly marker and crayon drawing: crooked proportions, scribbly lines, intentionally bad meme energy.

Proving identity in a "low-quality" style is difficult—and that's exactly where face lock shines.

Explicitly request "NOT realistic, NOT polished". Deliberate chaos.

Skin shop card: blue gradient background, rarity banner, fictional V-Bucks prices, full-body render.

Interface layout + recognizable character = collectible. Don’t copy text from a real brand.

Fictional cosmetic icons + polished stylized realism.

Anime/concept art champion poster with original creatures around the subject and a heroic three-quarter pose.

Combines semi-realistic painting, a Japanese magazine cover, and premium trading-card art in a single prompt.

Original creatures (do not copy IP) + minimalist cream background.

8K portrait in the style of open-world promotional art: neon, wet asphalt, luxury car, calm attitude.

Shows cinematic grading and poster typography — an AAA game cover look.

Inspired by the style, without using a real brand. Fictional typography and seals.

Superhero editorial portrait: metallic black suit with geometric patterns, helmet at the side, giant mechanical zord behind.

Here, Ref 2 inspires the fashion, but the outfit is transformed into a new concept (black, not blue).

Low angle + golden hour + minimalist futuristic HUD.

Plasticine caricature: noodle-like limbs, oversized hands, misaligned eyes, handmade clay texture.

Stop-motion style tests the limits of recognizability—exaggerated features, but still “the person.”

Miniature setting + macro depth of field + outfit recreated as clay fabric.

Hand-drawn ID card: crooked text, stickers, fake stamps, portrait with a goofy smile.

Document + doodle mix: structured layout with deliberately chaotic execution.

Fictional, silly details. Must NOT look like a real document.

Grotesque fashion caricature: stretched anatomy, dramatic pose, handwritten annotations, and fake fashion ratings.

"Editorial analysis" page from a strange sketchbook—exaggeration with purpose.

Messy ink + watercolor + aged beige paper.

Designer toy portrait: soft Pixar-style CGI + street fashion vinyl, glossy accessories, patches, and badges.

Wraps up the pack with a premium, futuristic, cute collectible style.

Subsurface scattering + product photography lighting + studio gradient.

A second take on the clay figure, showing that the identity lock holds up to extreme exaggeration.

Variations within the same genre expand the set without creating prompts from scratch.

Reuse the style language, change the pose and props.

Fake ID humor: another way to show that document and doodle can coexist.

Shows how rigid layout and chaotic execution create memorable results.

Strong identity survives even in a “bad on purpose” style.

Grotesque fashion sketch language pushed to the extreme, with sarcastic annotations.

Confirms that "sketchbook" styles work when the identity lock is strong enough.

Strange haute couture silhouette + fabric callouts.

The face comes from one reference and the fashion from another — so the result is recognizable without being locked into a pose.

It’s the principle you can carry over to any new style you come up with.

Fixed identity + flexible styling = repeatability.

Oil painting, doodle, game render, anime poster, open-world promo, sentai editorial, clay, fake ID, sketch, and CGI toy.

Clearly defined visual genres produce very different outputs of the same person.

The more specific the genre, the more distinctive the result.

The step-by-step process for generating: choose a model, lock in the identity, describe the world, and iterate on the pose.

From concept to practice—including how the images in this guide were made locally.

An edit-capable model for 2 refs; T2I for a single style. Iterate cheaply, finalize at high quality.