Back to blog
Nano BananaAI ImageCharacter Consistency

Nano Banana Character Consistency Prompts for Multi-Scene Work

Use Nano Banana character consistency prompts to preserve facial identity, hair, wardrobe, and signature details while changing one scene variable at a time.

By Genzy Editorial Team 4 min read
Nano Banana Character Consistency Prompts for Multi-Scene Work visual guide

Ready to test the method? Build your next text-to-image prompt.

Character consistency improves when the prompt separates identity from the scene. Nano Banana should receive a short list of features that remain fixed, a clear description of the one change you want, and a reference image where those features are easy to inspect.

The prompt patterns below are designed for Nano Banana editing and reference workflows. Availability in Genzy depends on which image models are enabled for the workspace.

Build one clean identity anchor

Choose a reference with a readable face, natural perspective, visible hairstyle, and important wardrobe details. Flat or soft light is easier to carry into new scenes than heavy colored lighting. Avoid using a distant group shot, motion blur, hands covering the face, or an extreme profile as the only identity source.

Before expanding the character, create a second approved view—a three-quarter portrait or medium shot. Two clear views reveal more information about face shape, hair silhouette, costume construction, and accessories.

Write a compact identity lock

List five to eight stable traits. For example: “Preserve the same adult woman: oval face, wide-set dark brown eyes, straight brows, small mole below the left eye, chin-length black bob with blunt fringe, silver ear cuff, rust-red field jacket.”

Choose traits that distinguish the person. Generic words such as beautiful, cinematic, stylish, or detailed do not help recognition. Do not repeatedly restate features that the reference already shows unless they are essential and prone to drift.

Separate the fixed block from the change block

Use two explicit sentences: “Keep the character’s facial identity, age, hairstyle, ear cuff, jacket construction, and body proportions unchanged. Change only the location to an elevated train platform at blue hour; medium-wide framing, cool overhead light, wet floor reflections.”

The phrase “change only” is useful because it defines the edit boundary. It does not guarantee perfection, but it makes the requested revision easier to evaluate.

Change one axis at a time

An axis can be pose, expression, camera angle, wardrobe, location, lighting, or art style. Changing all of them in one generation makes it difficult to identify what caused identity drift. Move from front portrait to three-quarter portrait, then to medium shot, then to a wider action scene.

Each approved image becomes another reference for the next step. Over time you build coverage of the character from multiple angles instead of asking one portrait to explain everything.

Copy-ready Nano Banana prompt patterns

For a new expression: “Preserve the exact facial identity, age, hairstyle, skin details, ear cuff, and jacket from the reference. Change only the expression to a restrained smile with relaxed eyes. Keep the same camera angle, crop, lighting, and background.”

For a new location: “Use the same character from the reference with unchanged facial structure, bob haircut, silver ear cuff, rust-red jacket, and body proportions. Place her in a quiet modern library beside a tall window. Medium shot, natural overcast light, realistic perspective. Do not redesign the character or clothing.”

For a new camera angle: “Preserve the person and costume exactly. Rotate the camera to a natural three-quarter left view while keeping a medium-close crop and similar soft light. Maintain eye spacing, nose shape, jawline, hair length, fringe shape, ear cuff, and jacket collar.”

For a wardrobe variation: “Keep facial identity, age, hairstyle, ear cuff, body proportions, camera angle, and lighting unchanged. Replace only the rust-red field jacket with a navy wool coat of the same length. Keep the underlying black shirt and all other details.”

Protect signature details without overconstraining

Select one or two recognition anchors beyond the face: a specific fringe shape, scar, earring, glasses frame, jacket collar, or color pair. If the prompt locks twenty tiny details, the model may satisfy some while neglecting the features viewers actually use to recognize the character.

Prioritize the face, hair silhouette, age, body proportions, and signature accessory. Treat small fabric texture or minor background objects as flexible unless continuity requires them.

Use a continuity sheet

Place approved images in a grid and compare them at thumbnail size and full size. At thumbnail size, check hair silhouette, head-to-body proportion, clothing shape, and color. At full size, check eye spacing, jaw, nose, hairline, accessories, garment seams, and repeated props.

Reject a beautiful image if it weakens identity enough to confuse the sequence. An approved continuity set should look like one character photographed on different production days, not several similar people sharing a costume.

Prepare the character for animation

Choose the most stable image as the first frame for image-to-video. Begin with breathing, blinking, a small head turn, fabric response, or a restrained camera move. A dramatic spin or full-body action forces the video model to invent unseen views and can expose inconsistencies that were hidden in the still.

Keep a written identity card beside the approved reference set. When the scene changes, reuse the fixed block and revise only the change block. This simple structure turns Nano Banana character consistency from repeated guesswork into a reviewable production workflow.