How to Prompt Nano Banana 2.1

You can picture the image. How do you turn it into instructions Nano Banana 2.1 can follow?

We’ll start with an underwater fashion photograph and build its prompt one part at a time. Then we’ll move into reference images, consistent characters, readable text, and edits that keep your design intact.

How to Write Your First Nano Banana 2.1 PromptCopy Anchor Link

Let’s create a fashion editorial: a woman in a flowing crimson gown suspended underwater, photographed from below as sunlight breaks through the surface.

You could write:

A woman in a red dress underwater. Beautiful, cinematic, dreamy.

But several important decisions are missing. Does the gown hang or spread around her? Is she swimming or suspended? Is the water an ocean full of life or a clean expanse of blue? Should the image look photographed or illustrated?

We’ll answer those questions in the prompt. The method is natural language: describe the subject, setting, action, composition, and style.

Describe the Subject and Its PositionCopy Anchor Link

Start with the part that makes the idea work:

A woman in a long crimson silk gown floats suspended in clear turquoise water. The gown spreads weightlessly around her, wider than she is tall, and her arms rest at her sides.

“Suspended” fixes her pose: she is not swimming. “Spreads weightlessly” tells the fabric to behave like underwater cloth, not draped cloth. “Wider than she is tall” gives the gown a scale relative to her body. Those instructions matter more to this concept than adding another word for image quality.

For your own image, describe the relationship between its main objects. A flower growing through a keyboard is different from a flower lying beside one.

Add the Setting and Camera AngleCopy Anchor Link

Now decide where she sits and how we see her:

Photograph her from slightly below, looking up toward the bright surface, with the sun breaking through as visible rays around her silhouette. Keep her full figure and the entire sweep of the gown visible, with deep blue water filling the rest of the frame.

Looking up lets the surface light silhouette her, a view only this setting offers. A level or downward angle would lose the rays and turn her into a swimmer, while a tighter crop would lose the gown’s sweep that carries the image.

The deep water around her also has a purpose: the magazine can add a headline later. It does not ask the model to invent that headline now.

Choose the Lighting and MaterialsCopy Anchor Link

We want the photograph to look like a camera actually caught this underwater:

Let the surface sunlight backlight her hair and outline the gown’s edges. Show the fabric’s translucency, tiny bubbles rising from its folds, and soft caustic light rippling across her skin.

Backlight gives the figure an edge. Even underwater light would flatten her into the blue background. The instruction chooses a visible result rather than asking for “dreamy lighting.”

Material details complete the illusion. Translucent silk, skin, and water should not look like they are made from the same smooth substance.

The Complete Text-to-Image PromptCopy Anchor Link

Inputs: None. Suggested setup: 16:9 landscape, 2K output.

Create a wide editorial fashion photograph, in 16:9.

A woman in a long crimson silk gown floats suspended in clear turquoise water, ten metres below the surface. The gown spreads weightlessly around her, wider than she is tall, and her arms rest at her sides. Her face is calm and directed slightly toward the light above.

Photograph her from slightly below, looking up toward the bright surface, with the sun breaking through as visible rays around her silhouette. Keep her full figure and the entire sweep of the gown visible, with deep blue water filling the rest of the frame.

Let the surface sunlight backlight her hair and outline the gown’s edges. Show the fabric’s translucency, tiny bubbles rising from its folds, and soft caustic light rippling across her skin. Include only the figure, the gown, and the water, with no lettering.

Result

Read it back as a set of decisions: what appears, how the objects relate, where they sit, how we see them, and how they are lit.

You can write these decisions in one paragraph. The breaks simply make a longer prompt easier to revise.

Nano Banana 2.1 supports 1K, 2K, and 4K output. Choose resolution and aspect ratio in the dashboard's output controls; the prompt describes the composition within that canvas.

How to Control Framing, Lighting and StyleCopy Anchor Link

A useful image can still need a different crop, softer shadows, or another material treatment. These are separate decisions. Choose the one you want to change before asking for a new version.

The next three lessons use different subjects. Within each lesson, keep the subject steady so you can see what the instruction changes.

How to Choose Framing for the Story You Want to ShowCopy Anchor Link

Imagine a fashion photograph of a woman holding an enormous silver crescent balloon in an empty concrete courtyard. Is the picture about her expression, the outfit, or the scale of the balloon?

Your answer determines what must fit inside the image. Framing is how much you show; camera angle is where you look from. A close portrait and a wide environmental view can both be photographed at eye level.

For this example, the oversized balloon is the idea. A tight face crop would remove the comparison that makes it interesting. Choose the third view, then add a camera position:

Keep the camera at the woman's eye level. Place her below the centre, with clear space between the balloon and the top edge.

Complete prompt. Inputs: none. Suggested canvas: 4:5.

Create a fashion photograph of an adult woman in a floor-length cobalt coat holding the thin tether of an enormous silver crescent-shaped helium balloon.

Place her in an empty pale-concrete courtyard. The balloon floats just above her and is taller than her upper body. Use a wide view showing the complete woman, her shoes, the full balloon, and the surrounding courtyard. Keep the camera at her eye level, with clear space above the balloon and below her feet.

Use soft overcast daylight and restrained colours. Show the balloon as reflective foil with a visible seam, and the coat as heavy wool. Keep the background empty and include no lettering.

Wide view

Full-body crop

Close portrait

The view is successful for this brief when the person supplies a clear sense of scale. If the balloon is cut off, correct the framing instruction before adding lens names. For more visual definitions, see our camera shot and framing guide.

Change the Light Before Rewriting the Whole PromptCopy Anchor Link

Suppose you like a fashion portrait's face, pose, and colours, but its shadows look too severe. You need to describe the lighting change without asking for a different portrait.

We will use a woman wearing a giant pleated-paper collar. Its folds make the shadows easy to compare.

Decide What You Want the Light to ChangeCopy Anchor Link

“Better lighting” leaves the model to choose the problem. Start with what you can actually see:

For our portrait, choose softness. Keep direction, colour, and overall brightness similar. Otherwise, several changes would be happening at once.

Understand Hard Light and Soft Light Before Choosing OneCopy Anchor Link

Look at the border where a shadow meets the lit surface. A distinct edge is the characteristic we call hard. A gradual transition is what we call soft. The apparent size of the light source matters: a source that looks larger from the subject produces softer shadow transitions.

For this portrait, hard light will be used to request strong strips of light and shade across the folded collar. Soft light will be used to request gentler transitions while retaining the folds.

Soft light does not mean a blurry photograph. Keep the eyes and paper edges in focus. Nor does choosing hard light require making the whole picture dark. Brightness and shadow softness are different decisions.

Create the Starting Portrait with Hard LightCopy Anchor Link

Inputs: none. Suggested canvas: 4:5. Generate this image first and save a version whose face, pose, and collar you want to retain.

Create a chest-up fashion portrait of an adult woman wearing an enormous butter-yellow pleated-paper collar that fans out behind her head. Keep her face and neck unobstructed. She wears a simple charcoal dress, with dark hair slicked back and a calm expression facing the camera.

Use a plain muted-lilac background and an eye-level view. Show the collar's complete fan shape, its thin paper edges, and precise accordion folds.

Light her with one small, hard, neutral-white source above and to the viewer's left. Make the alternating lit and shaded areas across the folds distinct, with a clearly edged collar shadow falling to the right on the background. Retain detail on the shaded side of her face. Keep eyes, skin texture, and paper edges sharply focused. No text or extra props.

Build the Soft-Light Edit One Decision at a TimeCopy Anchor Link

Upload the portrait you approved. Begin with the operation:

Change only the lighting in this portrait.

Then choose the source and keep its direction:

Replace the small hard light with a large, softly diffused neutral-white light from the same upper-left direction.

“Large” describes the requested source. “Same upper-left direction” prevents this edit from also becoming a front-lit portrait.

Now describe where the change should be visible:

Soften the shadow edges across the face and between the paper folds. Let the collar's shadow fade gradually into the background.

Finally, name the details outside this edit:

Keep the face, expression, collar shape, framing, and colours unchanged. Retain sharp eyes and crisp paper texture.

Complete edit prompt. Input: the approved hard-light portrait.

Change only the lighting in this portrait. Replace the small hard light with a large, softly diffused neutral-white light from the same upper-left direction.

Soften the shadow edges across the face and between the paper folds. Let the collar's shadow fade gradually into the lilac background while keeping the folds visibly three-dimensional.

Preserve the woman's identity, expression, hair, dress, collar construction, camera angle, crop, and colours. Keep overall brightness comparable to the original. Retain sharp eyes, natural skin texture, and crisp paper edges.

Hard-light portrait

Soft-light edit

Compare the shadow edges, not just which image you prefer. The requested change is a softer transition, with the same collar and face. A different collar, blurred skin, or a newly frontal light is a separate change to correct.

For a new text-to-image generation, replace the original lighting sentence with the soft-light instruction. For an image you already like, use the image-editing route above. Do not leave “small hard light” and “large soft light” in the same role in one prompt.

For a deeper treatment of light quality and direction, see our lighting prompt guide.

How to Change Visual Style Without Losing the ActionCopy Anchor Link

For style transfer, decide which information belongs to the subject and which belongs to the medium. A photograph of a runner can become embroidery while retaining the lifted knee and separated arms.

Prepare: a full-body runner photograph with the shoes and spaces between the limbs visible.

The pose makes the action recognizable. Keep it. The skin, fabric, and photographic shading will become thread. Change those. For this exercise, omit the original setting so the result reads as a patch rather than an embroidered landscape.

Build the instruction around those decisions:

Preserve the runner's pose and clothing colours. Rebuild the figure from directional thread stitches on navy fabric.

Then describe the physical signs of the new medium: visible weave, raised stitches, and a stitched border. These make “embroidery” more specific than “handmade style.”

Reinterpret the runner in the attached photograph as an embroidered patch on dark navy fabric.

Preserve the full-body pose, clothing colours, and the spaces between the arms, torso, and legs. Replace photographic surfaces with directional thread stitches, using finer stitches around the face and thicker stitches across the clothing.

Frame the figure with a thin stitched oval, leaving a gap where the raised shoe crosses it. Show the patch straight-on, with visible fabric weave and slight thread relief. Include the runner and oval only, without the original background or lettering.

The comparison should show the same action in a different material. If a shoe moves or an arm changes position, revise the pose-preservation instruction. Adding more thread adjectives would not address that mismatch.

Source photograph

Embroidered result

How to Use Reference Images in Nano Banana 2.1Copy Anchor Link

A reference gives the model visual information you do not have to reinvent in words. It also contains details you might not want. Assign its role before describing the new scene. Nano Banana 2.1 accepts up to 14 reference images per request; the examples here use only the inputs their task needs.

Use the actual upload labels in your interface when a prompt says “Image 1.” Prepare your own images or material you have permission to use. Our guide to selecting the ideal input images covers what makes a strong reference.

How to Preserve a Product While Changing Its BackgroundCopy Anchor Link

Our product image is a compact silver camera presented on a dinner plate. The creative setting can change; the product's design cannot.

Prepare: a clear reference showing the front and top of the camera and its wrist strap.

First decide what must match the product: lens shape, controls, proportions, finish, and branding. Then decide what belongs to the new photograph: plate, tablecloth, shadows, and reflections.

This distinction resolves a common conflict. “Keep the camera exactly the same” is useful for its design, but its old studio reflections should not remain when the new scene uses warm overhead light.

Use Image 1 for the compact camera's product design. Create an editorial photograph presenting it as the main course on a white dinner plate over a deep burgundy tablecloth.

Preserve the lens, controls, proportions, silver finish, wrist strap, and existing branding. Use a three-quarter view showing the front and top. Coil the black strap beside the camera on the plate, with one polished fork outside the plate on the left.

Light the scene from a warm overhead source. Let the camera's reflections and the shadow beneath it match the plate's lighting. Leave the surrounding table empty. No added copy.

Inspect the controls and lens, not just the overall silhouette. If a side panel matters but is hidden in the input, supply another view. A new angle cannot reveal exact source information that was never provided.

Image 1: product reference

Result

How to Combine Multiple Reference Images into One SceneCopy Anchor Link

Imagine an opened sardine tin containing a miniature sleeper suite, with a woman reclining inside and reading. We have separate photographs of the guest, the tin, and a train interior. None shows the finished scene.

Your prompt needs to do two things: assign a job to each image, then explain how those pieces fit together.

Prepare: a full-body photograph of an adult woman in rust satin pyjamas; a clear view of an empty opened silver tin; and an interior photograph showing a blue upholstered sleeper berth, bedding, cushion, and reading lamp. Keep these as three separate uploads.

Assign Each Reference a Specific JobCopy Anchor Link

The guest photograph contains a pose and studio background as well as a person. The interior photograph contains a whole cabin as well as furniture. Choose which details belong in the new picture:

Begin by naming those roles. Do not rely on the order of the uploads to explain them:

Image 1 supplies the guest and her clothes. Image 2 supplies the outer container. Image 3 supplies the furnishings inside it.

Explain How the References Fit TogetherCopy Anchor Link

The role list identifies the ingredients. It still does not say whether the tin goes inside the train, sits beside the woman, or contains the entire miniature room.

Make the relationship explicit:

Build the sleeper suite inside the opened tin. Keep the tin as the outer structure, with its metal rim and curled lid visible.

Then resolve the scale:

Scale the woman and furnishings together so she can recline comfortably inside the tin.

Scaling only the woman would leave the furniture too large. Scaling the guest and furnishings together gives the miniature a consistent size relationship.

Finally, choose a view that lets the reader understand the construction. An elevated three-quarter view can show the guest, the furnished interior, and the outside of the tin together. A close crop of the woman's face would hide the idea.

Copy the Complete Multi-Reference PromptCopy Anchor Link

Upload order: guest, tin, sleeper-suite interior. Suggested canvas: 4:5.

Create a surreal editorial travel photograph of a miniature sleeper suite built inside the opened sardine tin from Image 2.

Image 1 defines the woman, including her face, haircut, rust satin pyjamas, and bare feet. Image 2 defines the container's exact silver outline, rounded corners, rolled rim, embossed side ribs, and attached curled lid. Image 3 supplies the interior's blue quilted upholstery, ivory linen, apricot cushion, pleated lampshade, and brass details.

Fit one blue upholstered berth inside the tin. Scale the woman and furnishings together so she fits comfortably within it. She reclines against the cushion with her legs extending along the berth, holding a small open book and looking up towards the camera. Show her face clearly.

Keep the metal rim visible around the interior and the curled lid exposed at the far end. Mount a miniature reading lamp against the inner wall. The tin remains the outer structure; the train reference contributes furnishings rather than a second room around it.

Place the tin diagonally on a matte dusty-apricot tabletop. Photograph from above at a three-quarter angle, with the complete tin filling most of the frame. Use soft daylight and a restrained warm glow from the reading lamp.

Render it as an intricately built physical miniature: fine metal edges, real upholstery, satin folds, and believable contact shadows. One guest, one tin, one berth. No lettering or surrounding props.

Inspect three things separately: does the guest match Image 1, does the container retain Image 2's shape, and do the furnishings carry Image 3's design? Then check the relationship: the suite belongs inside the tin, and the woman fits inside the suite.

Image 1: guest

Image 2: opened tin

Image 3: suite interior

Result

The guest, metal container, and interior furnishings come from three different references.

How to Use a Style Reference Without Copying Its SubjectCopy Anchor Link

This lesson uses two new images: a photograph of a dancer leaping above an office chair, and a two-colour moth print. We want to redraw the dancer and chair in the print's style. We do not want a moth in the result.

The earlier embroidery example described a medium in words. Here, a second image supplies the visual treatment. The main decision is which parts of that reference count as style.

Prepare: Image 1, the complete dancer-and-chair photograph with clear space above the figure; Image 2, a flat moth print using blue and red ink on pale butter-coloured paper.

Separate the Subject from the Printing StyleCopy Anchor Link

A request to “make Image 1 look like Image 2” leaves the pose, subject, colours, and arrangement open to interpretation. Divide them instead:

Dot-pattern shading, also called halftone, builds lighter and darker areas from dots. A small offset between colours makes the printed layers look slightly misaligned. These are visible features you can point to in the reference, rather than a general request for a vintage look.

Write the division plainly:

Image 1 controls the dancer, chair, pose, and arrangement. Image 2 controls the ink colours, dot shading, and printing texture.

The charcoal suit will become blue and red ink shapes. Keep its silhouette and folds recognizable, but let its photographic colours change. Asking to preserve every original colour would conflict with the two-colour treatment.

Copy the Complete Style-Reference PromptCopy Anchor Link

Upload order: dancer photograph, moth print. Suggested canvas: 4:5.

Create a flat two-colour editorial art print.

Use Image 1 for the dancer's identity, exact airborne pose, suit, trailing tie, office chair, and their positions within the composition.

Use Image 2 only for its visual treatment: deep ultramarine and muted vermilion ink on pale butter paper, broad printed colour shapes, coarse dot-pattern shading, subtly uneven ink coverage, and slightly offset colour layers.

Rebuild the dancer and chair in that printing style. Translate their photographic light and shade into blue and red ink shapes, retaining the recognizable face, clothing silhouette, and clear spaces between the limbs. Keep the chair beneath the dancer in its original position.

The dancer and chair are the only pictorial subjects. The moth from Image 2 is a style reference, not an element to include. Preserve Image 1's action and arrangement rather than adopting the moth's symmetrical layout.

Keep the clear space above the figure. Include no lettering, decorative frames, printing marks, or extra graphics.

Compare the source photograph with the result first: the leap, gaps between the limbs, and chair position should remain recognizable. Then compare the rendering with the moth print: the palette and printed texture should transfer without its insect or symmetrical composition.

Keep this first result text-free. Adding a headline at the same time would introduce a separate layout task before you have checked the style transfer.

Image 1: dancer photograph

Image 2: moth print

Result: OFF DUTY variant

The photograph supplies the action and composition. The moth print supplies the ink colours and printing texture, not the subject.

How to Keep a Character Consistent Across ImagesCopy Anchor Link

Keeping the same character means retaining identity while allowing a new expression, pose, and view. Treat those as separate decisions.

For this lesson, use a new character: Mika, an adult bicycle courier outside a flower shop at blue hour. Prepare a front portrait, a profile portrait, and an approved outfit image showing an off-white jacket with a plum collar and a red crossbody bag.

How to Change Camera Angle Without Freezing the Original PoseCopy Anchor Link

A front portrait is not a command to face the lens in every image. Tell the model what the reference establishes, then specify the new head direction and camera position.

Images 1 and 2 establish Mika's identity. Show her looking towards the right edge of the picture, with the camera beside her so her face appears in profile.

“Looking right” sets her eyeline. “Camera beside her” sets the view. Asking for one does not automatically specify the other.

Create a side-profile photograph of Mika, the adult courier in Images 1 and 2. Use Image 3 for her off-white jacket with a plum collar and red crossbody bag.

At blue hour, she stands beside her bicycle outside a closed flower shop. A wrapped bunch of sunflowers rests on the front rack. She lowers her head slightly to read a folded delivery note.

Place the camera beside her at eye level, showing her profile as she looks towards the right side of the picture. Preserve her facial identity, haircut, age, and outfit design. Allow the expression, clothing folds, and shadows to respond naturally to the new pose.

Use warm shop-window light against the cooler street. Keep her face readable and include no added lettering.

Compare the face with the profile reference and the outfit with its own image. Do not judge identity by whether Mika repeats the front portrait's expression.

Image 1: front portrait

Image 2: profile portrait

Image 3: outfit reference

Result: profile scene

How to Plan a Consistent Image SeriesCopy Anchor Link

Separate the information that lasts through the story from the action that changes in each picture.

For the next scene, reattach the identity and outfit references and use the approved flower-shop image as location context. Give the new image one event, for example Mika lifting the flowers from the rack. Do not describe the whole delivery in one still-image prompt.

Approve each image before using it as context for the next. If a jacket pocket changes, return to the outfit reference rather than treating that accidental change as the new design.

How to Prompt Multiple Characters Without Mixing Faces or ActionsCopy Anchor Link

Several faces in one image create several separate instructions. A list of names is not enough; connect each name to a reference, a position, and an action.

Our scene is three dinner guests competing for the final strawberry. Prepare three adult character references: Ada in a plum suit, Bo in a navy sweater, and Cy in pale yellow.

Assign Each Person a Visible Role Before Writing the SceneCopy Anchor Link

Write the roles as a small production plan:

Now choose one moment. In this image, Ada's fork has not yet touched the strawberry. If you also ask everyone to eat it, the brief contains competing moments.

Create a deadpan editorial photograph of exactly three adult dinner guests competing for the last strawberry.

Ada from Image 1 sits on the viewer's left in her plum suit, reaching a fork towards one strawberry on a white plate at the centre of the table. She looks at the strawberry. Bo from Image 2 sits on the viewer's right in his navy sweater, holding the plate's rim and looking at Ada. Cy from Image 3 sits between them, farther back in the pale-yellow shirt, holding an empty fork upright and looking into the camera.

Keep each face, hair, and outfit associated with its own reference. There is one of each guest, one plate, one strawberry, and two forks. Capture the instant before Ada's fork touches the fruit.

Use a mint diner booth, a reflective tabletop, and direct flash. Show all three faces and the hands carrying out the actions. Include no text.

Review the relationships before the decoration: who holds the plate, who reaches, who looks at us? A beautiful diner cannot make a swapped role correct.

Use “viewer's left” to place someone within the image. Use “Ada's left hand” only when the anatomical side matters. Those are different reference points.

Image 1: Ada

Image 2: Bo

Image 3: Cy

Result: dinner scene

How to Create Readable Text, Posters and InfographicsCopy Anchor Link

Text introduces a new requirement: the picture has to contain the right words in a usable arrangement. Decide the wording before the font treatment. For lettering fundamentals, see our text-in-image guide.

How to Plan Poster Typography Before Generating the ImageCopy Anchor Link

For a fictional film poster called NIGHT SHIFT, choose what the viewer should notice first. We want a large title, then a face, then a small date. This reading order is the visual hierarchy.

Start with a text plan rather than “beautiful typography”:

Next choose a lettering contrast. Narrow upright letters can fit the first word across the top; an italic style can make the second word feel different. Condensed means narrow. Serif refers to the small finishing strokes on the letters. These terms describe the appearance you request, not a guaranteed match to a font file.

Finally, decide what sits in front. The hair may overlap a small part of the lettering. The letters must not run across the eyes. A request to “blend text with the person” does not explain that relationship.

Prepare: one adult portrait. Suggested canvas: 4:5.

Create a flat editorial poster for a fictional short film titled "NIGHT SHIFT". Use Image 1 for the performer's face and hair.

Show the performer in an ink-blue courier jacket outside a closed convenience store at night, looking directly at the camera and holding a coral-red bicycle helmet against one hip. Use muted olive and near-black surroundings with soft shop light on the face.

Set "NIGHT" across the top in very large cream condensed sans-serif letters. Put "SHIFT" beneath it in large cream italic serif lettering. Let the hair overlap a small part of the title's lower edge, while preserving readable letter shapes and leaving the eyes clear.

Put "25 OCT" in one small coral circle in the lower-left background, clear of the body. Include only the title and date. Deliver the flat poster artwork, not a photograph of a poster on a wall.

Check the title at the intended display size. If it disappears into the shop window, address contrast behind that title before adding outlines, glowing edges, and additional labels. For example:

Darken the window directly behind the cream title. Keep the face exposure, letter shapes, and all other poster elements unchanged.

That is a background change serving legibility. It is not a request to restyle the whole poster.

Image 1: portrait

Result: poster

How to Make an Infographic Without Changing the Meaning of the DataCopy Anchor Link

An infographic needs a rule connecting its shapes to its numbers. We will show ten fictional working hours as paper tickets. One ticket means thirty minutes.

Work out the graphic before asking for a design:

Twenty tickets fit a four-column, five-row grid. Every ticket must have the same size because every ticket represents the same amount of time. Colour distinguishes categories; size does not.

Now add the visual treatment: punched paper edges, muted colours, and clear labels.

Create a 4:5 editorial infographic titled "WHERE 10 HOURS WENT" using this fictional dataset only.

Show exactly 20 equal-sized paper tickets in four columns and five rows. One ticket represents 30 minutes. Reading left to right and then top to bottom, colour the first 8 muted plum, the next 6 sage green, the next 4 ochre, and the last 2 pale grey. Give each ticket matching dimensions and punched semicircles on the short edges.

Below the grid, write:
"Design: 4 hours"
"Calls: 3 hours"
"Admin: 2 hours"
"Learning: 1 hour"

Match each category label to its ticket colour. Add "1 ticket = 30 minutes" and "Illustrative data" beneath the labels. Use dark readable lettering on warm white. Include no additional figures or claims.

Count the units in the output. A twenty-first ticket is an information error, even if it fills the grid attractively. Check the words and numbers separately from the artwork.

Result

When Should You Increase the Thinking Level?Copy Anchor Link

Nano Banana 2.1 supports minimal, medium, and high thinking; medium is its default. Treat the setting as a way to allocate reasoning effort, not as a style preset. For PhotoGPT's other output controls, see advanced settings for image generation.

For the ticket infographic, several conditions depend on one another: the total, grid, colour groups, and labels. High is a reasonable setting to try. For a single subject against a simple background, start with less elaborate planning and inspect the result.

Do not change the prompt, resolution, and thinking level together when trying to understand a particular improvement. Keep the brief and output size steady. Inspect whether the specific problem, for example the wrong ticket count, changed. A more elaborate picture alone does not demonstrate more accurate reasoning.

How to Edit an Image Without Redesigning ItCopy Anchor Link

An editing request has a base image and a target. Start from the base you actually approve, then describe the operation on that target. Nano Banana supports targeted and conversational editing.

How to Specify What Changes and What StaysCopy Anchor Link

For a new material exercise, prepare a photograph of a red rotary telephone on emerald-green velvet. The receiver rests on the cradle; its coiled cable is visible.

We want a milky glass receiver but a red telephone base. The word “telephone” refers to both parts, so “turn the telephone into glass” is too broad.

Build the request in three pieces:

Target: the receiver resting across the top.

Operation: replace its opaque red material with milky opal glass.

Preservation: retain the receiver's shape, the red base, the cable, and the arrangement.

Edit this photograph of the red rotary telephone. Change only the receiver resting across the top into milky opal glass.

Preserve the receiver's shape, earpiece and mouthpiece, size, and position. Give the glass soft translucent edges and highlights consistent with the existing light. Keep the telephone base red, the coiled cable black, and the velvet emerald green.

Retain the camera view, dial, object positions, and background. Let the receiver's reflections change with its new material, without making it glow from inside.

The new material needs different highlights. Asking for the identical red receiver reflections would conflict with the change. Protect the shape and placement; let the surface respond to its new material.

Before

After: opal glass receiver

How to Use a Mask or Marked Image for a Precise RegionCopy Anchor Link

Words can identify a receiver easily. They may struggle to identify one flower among many similar blooms. Use a marked reference when the target is visually ambiguous.

For this exercise, prepare a bouquet photograph and a copy with a blue outline around one flower head. The outline should exclude the stem and neighboring flowers.

State whether the mark means change here or protect this. Those are opposite instructions. Here it means change:

Image 1 is the original bouquet. In Image 2, the blue outline identifies the single flower head to change.

Turn that flower head into translucent cobalt-blue glass while preserving its petal arrangement and connection to the original green stem. Match the existing window light with bright glass edges and darker overlaps between petals.

Keep the other flowers, stems, leaves, vase, and framing unchanged. Use the outline as an instruction only; remove it from the final image.

The annotation selects the area. It does not specify the replacement material until the words do. A drawn outline is also different from a dedicated mask passed through an application's editing controls.

Check for changes outside the target and leftover coloured lines. Neither should be accepted merely because the intended flower became blue.

Original bouquet

Marked duplicate

Edited result

How to Write a Follow-Up That Fixes the Actual ProblemCopy Anchor Link

Suppose the glass flower now looks as though it contains a lamp. “More realistic” does not name that defect. Describe the correction:

Keep the cobalt glass petals and their shape. Remove the internal glow. Show illumination from the existing window, with small reflected highlights and darker areas where petals overlap. Keep the rest of the bouquet unchanged.

This asks for a different response to light, not a different flower. In a conversational interface, continue from the approved image. Otherwise, attach that version again. If a previous turn changed the wrong flower, go back to the earlier image rather than refining the mistake.

How to Translate and Resize an Existing DesignCopy Anchor Link

Translation changes the copy. Resizing changes the canvas. Neither tells the model how to rearrange the design automatically. Decide which elements may move before writing the request.

How to Translate a Poster Without Shrinking Its Main ImageCopy Anchor Link

Prepare a poster for an overnight flower kiosk with a large tulip photograph, the two-line headline “OPEN / ALL NIGHT,” and a small “24H” badge. This is a separate design exercise, not a claim about an existing business.

For the Spanish version, use the approved phrase “ABIERTO TODA LA NOCHE.” It is longer than the English headline. Choose three lines and permit a text-size adjustment. Do not require identical letter positions.

Create a Spanish-language version of this flower-kiosk poster.

Replace the existing "OPEN / ALL NIGHT" headline with:
"ABIERTO"
"TODA LA"
"NOCHE"

Use the same headline colour and lettering style. Keep the new copy within the original headline area, adjusting letter size and line spacing as needed. Preserve the tulip photograph, its scale and crop, the background, and the "24H" badge.

Keep clear space between the lettering and the flower. Fit the approved words before adding any decorative typography. Include no new copy.

The prompt supplies the translation instead of asking the model to invent campaign wording. That is useful when the copy has already been approved. Changing the language does not call for a different person, building, or cultural setting unless you request those changes too.

English: OPEN / ALL NIGHT

Spanish: ABIERTO TODA LA NOCHE

How to Choose Between Cropping, Extending and RecomposingCopy Anchor Link

For a wider output, identify the actual task:

For a square architecture poster featuring a spiral staircase and the title “TAKE THE LONG WAY,” a wide banner needs recomposition. Keeping every element in its square position would make the title and stairs unnecessarily small.

Prepare: the square poster. Select: 16:9 in the output controls.

Recompose this square architecture poster as a 16:9 website banner.

Place the staircase photograph in the right half, preserving its viewing angle and spiral geometry. Arrange the exact title "TAKE THE LONG WAY" in three large lines in the left half. Keep the original lettering style and palette, with generous space around the title.

Extend the existing wall background to connect the two areas. Adjust image placement, crop, and type size for the wider layout, but keep the stairs and letters at normal proportions. Do not invent another staircase or new copy.

Judge the new layout at banner size: can the title be read and the staircase recognized? Matching the original image's pixel arrangement would be the wrong goal for this task.

Square poster

16:9 banner

How to Turn a Sketch into a Product ConceptCopy Anchor Link

Prepare a side-view sketch of a chair formed from one continuous folded ribbon. Mark the seat and backrest clearly enough that their relationship can be seen without an explanation.

Decide what the sketch controls: silhouette, seat depth, backrest angle. Your prompt will add the material, setting, and finish.

Preserve the drawn structure. Render the chair as brushed olive-green metal.

That is more precise than “make this a realistic chair,” which leaves the construction open to redesign.

Turn the attached ribbon-chair sketch into a furniture concept photograph.

Keep the continuous folded silhouette, seat depth, and backrest angle from the drawing. Render the chair as a thick sheet of brushed olive-green metal with softly rounded edges. The folds form the chair itself, not decoration wrapped around another chair.

Keep the side view close to the sketch's viewing angle. Place the chair alone on a dusty rose floor against a matching wall. Use a hard upper-right light and show the full chair with its shadow. Omit drawing lines, dimensions, and labels.

Compare the silhouette against the sketch. A stylish chair with a different construction missed the structural brief. The output is a concept visualization, not an engineering validation.

Sketch

Rendered concept

Nano Banana 2.1 vs Nano Banana 2.0: What Should You Change in Your Prompts?Copy Anchor Link

Here, Nano Banana 2.0 means Nano Banana 2, the previous generation. Nano Banana 2.1 is the updated model.

The upgrade does not create a new command language. The useful distinction is between a setting that changed and a task it handles better.

Which Differences Require a Workflow Change?Copy Anchor Link

The settings and improvements above are general differences between the versions. They are not outcomes from the creative exercises here.

4K, references, and search grounding were already part of Nano Banana 2. Do not rewrite a working brief simply to make it look specific to the newer version.

How to Compare the Models on a Demanding Creative RevisionCopy Anchor Link

Use a task that contains competing requirements rather than asking each model for an unrelated attractive image. Photorealism is a demanding brief: the details that make a portrait believable are exactly the details a weaker generation or edit loses first.

Generate this portrait first and approve a version you want to keep:

Create a photorealistic close portrait of a retired fisherman inside a wooden cabin at dawn, 4:5.

An older man with weathered, sun-marked skin, short grey stubble, and faint lines around his eyes sits beside a rain-streaked window. He wears a heavy wool sweater and rimless spectacles. His gaze rests outside the frame, calm.

Light him with two sources that must agree: a warm oil lamp on the table at lower left, and cold blue dawn light through the window at upper right. Each raindrop on the glass catches the window light; his spectacle lenses carry a faint reflection of the lamp. Steam rises from a tin mug in the foreground.

At full zoom the image must hold skin pores, individual stubble hairs, fabric weave, water beads on the window, and the lamp's reflection in his lenses. Natural colour, no beauty smoothing, no text.

Every element above is a checkable stress point: pores and stubble versus waxy smoothing, two light sources whose shadows and catchlights must agree, spectacle lenses that must reflect the lamp, and raindrops that must refract the window light.

For a comparison, give each model the same starting image, prompt, resolution, and a supported thinking level. Then apply the same relighting edit, a global change with one fixed requirement: it must still be the same man.

The oil lamp has just burned out. Remove its warm light and its reflection in his spectacles; let only the cold dawn light through the window shape the scene, deepening the shadows on the lamp side. Keep his face, pose, clothing, the window, the raindrops, and the mug identical.

Nano Banana 2.0

Nano Banana 2.1

Inspect the edits at full zoom before judging which looks nicer. Check the same points on both: the lamp's reflection gone from the lenses, shadows deepened on the lamp side only, skin texture preserved rather than re-smoothed, and whether it is still the same man. A relight that rebuilds the face is a new image, not an edit.

Then test continued editing: use each model's own output for the next identical revision. This reveals whether an earlier change is retained across later turns. Keep that chain separate from the single-edit comparison; they answer different questions.

How to Fix a Prompt When the Result Misses the BriefCopy Anchor Link

Before another generation, identify which kind of mismatch you have. The correction should address that mismatch rather than add a general request for more quality.

Save the prompt, inputs, settings, and result together. When a correction works, that record tells you which instruction changed. When it does not, it stops you from mistaking a different-looking image for a solved problem.

SourcesCopy Anchor Link

• Google Cloud: Gemini Nano Banana 2.1

• Google AI for Developers: Gemini Nano Banana 2.1

• Google DeepMind: Nano Banana 2.1 model card