Nano Banana 2.1 prompts work hardest when they play to the model’s strengths: readable text on posters and diagrams, precise edits, changes directed by a doodle, scenes assembled from several images, and step-by-step refinement in one chat. The 14 prompts below are sorted by those jobs, so each one exercises something the new model does well. Every prompt runs in Picsart AI Playground, where Nano Banana 2.1 sits alongside the rest of the AI models catalog. Copy one, swap in your own subject, and generate.
What changes about prompting Nano Banana 2.1
The model comes from Google’s Gemini 3 family, with Gemini 3.6 Flash underneath. It takes words and pictures as input, and a reply can mix pictures with written text. As a result, a prompt can carry both a written brief and the photos it refers to. Google positions it for fast, precise rounds of creating and editing, legible lettering, world knowledge, and lettering in several languages. Resolution is set to 1K, 2K or 4K, and the frame can be anything from square 1:1 to an extreme 8:1 or 1:8 strip.
Google also publishes the weak spots, and a good prompt writes around them. Tiny lettering tends to blur at the lowest resolution, and long blocks of copy struggle, so the text prompts below keep wording short and call for 2K or 4K. Left and right occasionally get swapped, so these prompts anchor placement to objects in the scene instead. A character can drift a little from the uploaded reference, and an edited subject sometimes holds on to its old pose. For that reason, the character prompts spell out what stays and what moves.
Text and infographic prompts
Legible lettering ranks near the top of Google’s list of intended uses, and the new model outscored Nano Banana 2 on infographic design and factual accuracy. Pick 2K or 4K before running these, and set the frame to 8:1 for the banner.
A portrait poster for a neighborhood café’s autumn menu launch. Warm illustrated style with a steaming ceramic cup, cinnamon sticks and scattered maple leaves on a deep burgundy background. Render this headline exactly once, in large cream serif lettering across the top: “Autumn Menu”. Directly below it, in smaller cream lettering: “Menú de otoño”. Along the bottom edge, in small clean sans-serif: “From October 10”. Keep all text crisp and evenly spaced, with generous margins. No other text, no logos, no watermarks. A clean educational infographic titled “How a seed becomes a tree”, landscape orientation, white background. Show five stages connected by curved arrows: seed in soil, germination, seedling, young sapling, mature oak tree. Label each stage with one short phrase in dark green sans-serif text. Under the oak, add a small labeled detail of an acorn. Flat illustration, soft greens and browns, plenty of white space, every label large enough to read at a glance. An 8:1 website banner for a handmade ceramics studio. A long wooden shelf runs the full width of the frame, holding a row of glazed bowls, vases and mugs in sage, sand and speckled white. Soft morning light falls across the shelf from a window at the far end. Place the text “Made by hand, fired with care” in thin charcoal serif lettering on the plain wall above the center of the shelf. Calm, minimal, photographic.A bilingual café poster
An infographic that leans on world knowledge
A long banner for a website header
Editing prompts for objects, environments, styles and characters
Swapping objects, changing the setting, restyling a picture and adjusting a character all fall inside the editing range Google tested. Upload the image you want to change, then name one change and the things that must stay. Each of these prompts follows that pattern.
Replace the white ceramic vase on the dining table with a small glass bottle holding a single sprig of eucalyptus. Keep the table, the chairs, the window behind them, the lighting and the camera angle exactly as they are. Match the new bottle’s shadow and reflections to the existing light. Move this person and the bicycle they are holding from the city street to a quiet coastal road at golden hour, with low dunes and tall grass beside the road and the sea in the distance. Keep the person’s face, hair, clothing, pose and the bicycle exactly the same. Relight them with warm, low sunlight that matches the new scene. Restyle this photo of a mountain lake as a vintage travel poster: flat areas of color, simplified shapes, a limited palette of teal, ochre and cream, and subtle paper grain. Keep the composition, the shape of the mountains, the boat on the water and the line of pine trees exactly where they are. Add no text. Change only this character’s outfit to a yellow raincoat, dark green rubber boots and a striped scarf. Keep the face, hairstyle, expression, body shape and the background unchanged. Turn the character slightly toward the camera with one hand raised in a wave, so the new pose differs clearly from the original.Swap one object
Move the scene to a new environment
Restyle the whole image
Edit a character, keep the identity
Doodle edit prompts
Drawing on a picture is a way to steer an edit that Google evaluated directly. Mark the area that should change on the image itself, then describe the change in the prompt. Keep the instruction short and point it at the marks, because Google says the model does not yet follow every doodle instruction fully.
Inside the area circled in red, add a tall potted fiddle-leaf fig in a woven basket. Match the plant’s lighting and shadow to the rest of the room. Remove the red circle and leave everything outside it unchanged. Turn the rough white lines drawn on the sky into a flock of seagulls flying in the same arc as the lines. Keep the beach, the people and the sea exactly as they are, and remove the drawn lines completely.Fill the circled area
Follow the drawn shape
Multi-image prompts
Placing several products into a fresh scene and putting several characters in one frame are both part of what Google tested, and keeping multiple characters consistent is where 2.1 pulled furthest ahead of Nano Banana 2. Upload each image, then tell the model what every image is for.
Use the three uploaded product photos: a serum bottle, a jar of face cream and a lip balm tube. Arrange all three on a pale travertine bathroom shelf beside a folded linen towel and a small sprig of dried lavender. Keep each product’s shape, color, label and proportions exactly as in its photo. Soft diffused daylight from a frosted window, gentle shadows, square 1:1 frame for a product listing. Use the two uploaded character images: the orange cartoon fox and the small blue robot. Place them together on a park bench under an autumn tree, sharing a paper bag of roasted chestnuts. The fox sits closer to the tree trunk and the robot sits closer to the path. Keep both characters’ designs, colors and proportions exactly as in their images. Warm afternoon light, storybook illustration style matching the uploaded art.Several products in one styled scene
Two characters, one frame
A multi-turn chain
Editing across several turns of a chat is a supported mode, and Google’s test set included back-and-forth sessions. So instead of one long prompt, build the image in turns. Run these three in order in the same conversation, and check the result after each one.
A cozy reading corner in a sunlit apartment: a mustard yellow armchair, a tall brass floor lamp, a low stack of books on the floor and a large window with sheer curtains. Morning light, warm and soft, photographic, 4:5 frame. Keep the room, the furniture and the camera angle exactly the same. Change the view through the window to a snowy winter street, and add a knitted gray blanket draped over the arm of the chair. Keep everything from the previous image. Add a steaming mug of hot chocolate on the stack of books and a small string of warm fairy lights along the top of the window frame. Switch the lamp on so it glows softly. Name the subject in plain nouns, give one clear change at a time, and say what must stay the same. Also, keep any text short and quoted exactly. These habits suit the model’s strengths in precise editing and legible lettering, and they sidestep the weak spots Google publishes.Turn one: the base image
Turn two: change the season
Turn three: add the finishing detail
Get answers to common questions
Start prompting Nano Banana 2.1 in Picsart
The prompts above work because they ask Nano Banana 2.1 for the jobs it handles well, and they write around the weak spots Google publishes. Head to AI Playground, choose the model, and paste in the first one.
