SupportGet started

Create images

Generate AI images from a prompt, explore several directions at once, and choose the model, aspect ratio and resolution.

Last verified

Overview

Describe your image, choose a model and settings, then select Create. Results appear as they finish in your current project.

Generate several images per prompt to explore different directions for a cover, product shot or campaign.

Write a prompt

Type in the prompt box at the bottom of the page. Enter sends it and Shift+Enter starts a new line. For a long prompt, choose the expand button at the top right of the prompt box, or press ⌘⇧E (Ctrl+Shift+E), for more room to write. Pasting a long prompt opens it on its own, and Esc, a click outside it or sending shrinks it back.

Before your first image, Create shows a few styles to start from. Select one to add it to the prompt box, then describe what to make in it. See all opens every style.

Name the subject, setting and style: "A brown hare in an oversized olive coat walking along green hills at dawn, gouache, picture book style." Include the details that matter; the model fills in the rest.

Pick a model

Open the model menu from the composer. New accounts start on Nano Banana 2.1. The list shows names only. Point at a model to see its details beside the list: Speed, Quality, Cost per image, and whether it Supports working from your own images.

Some models have a setting of their own, shown below their details, such as Web search on Flux 3 Image or Magic Prompt on Ideogram 4.5. Not sure which to pick? Which model for what matches models to jobs, such as Ideogram 4.5 for posters and logos.

Choose the shape and size

Select the settings button in the composer, next to the model. It shows your current choice, such as 1:1 · 1K · ×4.

  • Aspect ratio sets the shape. Pick Portrait, Square or Landscape, then use the slider for an exact ratio, from 9:16 to 21:9. A model only offers the ratios it can make.
  • Resolution sets the size, on models that offer more than one. A few models call it Quality instead.
  • Images per prompt is how many images one press of Create makes: 1, 2, 4, 6 or 8.

Get more range

Images from one prompt on one model often look alike. For more range, turn on Vary prompts or use a variety pack. With either one on, the button changes to Batch Create and shows how many images it will make.

Vary prompts, at the bottom of the image settings, writes a different version of your prompt for each image. The first image keeps your prompt exactly as you wrote it. Turn it on when the prompt is right but the results look alike. Start with Auto; Customize lets you pick what changes, such as Style or Mood, and a Direction every version follows.

A variety pack runs your prompt on up to four models at once, giving you more creative directions in one run. Select + beside a model to add it, or double-click a row to return to one model. Each model produces the selected Images per prompt count: four models at 2 each makes 8 images.

One press makes up to 48 images, and each is priced like any other image from its model.

Find your results

Results are grouped by day. Each day starts with a heading that gives its name, such as Today, Yesterday or a date, beside its date or weekday. While you scroll through a day, its heading shrinks into a small label at the top of the page, so you always know which day you are looking at.

Select a day's name to jump to another day. The list groups days by month, and Find older work in Library at the bottom opens that day in the Library. The up and down arrows beside the name step to the newer or older day.

The runs you are making now are shown full size at the top of Today. Each shows its model, the prompt, any references and a line of details such as the aspect ratio and resolution, with the images beside it. A long prompt shows its first four lines; select it to read the whole prompt, copy it or use it again. When you point at the prompt, Copy and Use prompt appear at the end of its details line, so you can copy or reuse it in one click. The run's actions sit under its images and appear when you point at the images. They include Share, or Share all for a run of several images, which makes one link to the whole run. On a phone they are always shown, and the prompt comes after the images.

Older runs show their images as a wall. Each image keeps its shape, and a wider gap marks where one run ends and the next begins. When a run has more images than fit, the last tile shows how many more.

Point at an image in the wall to see its whole run outlined, on every line it spans. Select an image to open its run below that line of images, with every image full size, the whole prompt and the run's actions. The left and right arrow keys move to the previous and next run. Press Esc or select the close button to close it. The run you opened stays outlined in the wall. The expand button beside it switches to Rows, and the run's images move into their place in its row. In a run shown full size, select an image to open it in the lightbox.

Switch between Grid and Rows with the two buttons at the top right. Grid shows older runs as the wall, to scan a lot of work at once. Choosing Grid, even when it is already on, also folds the runs you are making now into the wall; runs you start after that show full size again. Rows shows every run full size, the way your current runs are shown, to read prompts and act on runs. What is at the top of the page stays there when you switch. You can also hold ⌘ (Ctrl on Windows) and scroll, which keeps the image under the pointer in place. Create remembers your choice.

Right-click an image for its menu, which includes Ask agent, Use as reference, Select, Add to references, Edit, Remove background, Describe, Copy image, Download, Share image, Favorite and Delete. Share image makes a link anyone can open; see Sharing.

The keyboard works on the image you last selected or moved to:

  • The arrow keys move between images.
  • Enter works like selecting the image, and Space opens it in the lightbox.
  • R reuses the run's prompt and settings, A asks the agent about the image, E opens it in the editor, T uses it as a reference, X selects it, and F favorites it.

Re-run has no key, because it spends credits right away. Find it in the arrow beside Reuse.

Look closer in the lightbox

Select an image in a run shown full size, or press Space on it, to open it in the lightbox. The arrow keys, or J and K, move through your images. Scrolling over the thumbnails beside the panel moves faster: each image shows as you pass it, and the full image loads where you stop.

The panel beside the image starts with the model and when it ran, with Favorite, Download and the ⋯ menu, which holds the same actions as right-clicking the image. Below them are the prompt, a line saying what each reference sent, such as "Sent with Brand look (4 of 12) and Mira", the style or references it used, and its details. A long prompt shows its first six lines, and Show all opens the rest. Copy copies the prompt, and Use prompt puts it in the prompt box.

At the bottom, Reuse loads the run's prompt and settings into the prompt box, with Re-run in the arrow beside it. Edit opens the image in the editor, and Ask agent puts it on a message to the agent. Point at a button to see its key: R reuses, E edits, A asks the agent and F favorites.

Click the image to zoom in on the spot you clicked, filling the window. Drag to move around it, or scroll. Pinch the trackpad, or hold ⌘ (Ctrl on Windows) and scroll, to zoom further in or out. The arrow keys move around, + and - zoom, 1 shows the image at its real size and 0 shows all of it. Click without dragging, pinch out past the whole image or press Esc to go back. Z zooms in and out from the keyboard. To close the lightbox, click the space around the image, select the close button or press Esc. On a phone, the image grows out of the tile you tapped. Swipe sideways to move between images, or drag the thumbnail strip at the bottom to move through them quickly. Pinch or double-tap to zoom in, and double-tap again to zoom out. Tap once to hide or show the controls. Swipe up, or tap the prompt, for the details. To close, pull the image down or select Close at the top, and it goes back into its tile.

Select several images

Hold ⌘ (Ctrl on Windows) and select an image to start a selection. In the grid you can also select the check that appears in an image's corner when you point at it. Once you are selecting, selecting an image adds or removes it, holding Shift adds every image between your last pick and this one, and dragging across images picks them all. Point at a run's model to see the check that selects the whole run, and while you are selecting, the check beside a day's name selects the whole day. ⌘A (Ctrl+A) selects everything loaded, and Esc clears the selection.

While images are selected, a bar above the prompt box shows how many, and acts on all of them:

  • Move sends their runs to another project, to No project, or to a new project you name with New project from selection…. A run's images always move together. Undo in the message that follows sends each run back where it came from.
  • Download saves one image as is, and several as a ZIP.
  • Favorite favorites them all, or unfavorites them when every one already is.
  • Ask agent puts up to eight of them on a message to the agent.
  • Share puts up to 200 of them on one link anyone can open.
  • Edit runs one edit on every image, such as removing the background, and the results land together as one stack.
  • Compare, with two to four images selected, shows them side by side.
  • Delete deletes the images you made. Images a teammate made stay, and the confirmation says how many.

The ⋯ button holds the rest: Use as references, Add to references, Re-run for runs selected whole, and Copy prompts. On a narrow window, Ask agent, Share, Edit and Compare move into it too.

What it costs

Check Cost in the model menu for the price per image. Higher resolutions and quality levels usually cost more. Multiply the price by your image count for the total. Open a finished image to see the cost of its set, or see Credit costs for all prices.

If you run out, the composer says so. On Free, the owner sees the smallest plan whose monthly credits pay for the run that was stopped (Pro when no run was stopped), with the welcome discount in your personal workspace while the offer is active. See plans opens Settings > Plan and credits with that plan picked and every plan beside it, and you pay through Stripe from there. If a run stopped because you ran out of credits, your prompt and settings are kept. After you pay, select Run your prompt to start it. On a paid plan the owner gets Add credits. A member sees who to ask instead.

Below 500 credits (150 on Free), the composer shows a low-balance notice with the same choice. Dismiss it and it comes back when the balance falls to the next hundred or runs out. Once you are out, dismissing it leaves one line above the composer until you have credits again.

If you left a plan's checkout without paying, the same place above the composer names the plan you picked, with See plans to open Settings > Plan and credits with it selected. It shows for 3 days, until you choose See plans, close it or subscribe, and gives way to the credit notices. See Plans.

Set your starting point

New prompts start with your default model, aspect ratio and images per prompt. Change them in the composer and the new choice becomes the default. You can also set them in Settings > Preferences, under Composer.

Tips

  • Find the idea on Nano Banana 2 Lite, at 8 credits an image. Once you know what you want, run the prompt again on Nano Banana 2.1 at 2K or 4K, or on Nano Banana Pro when the image needs fine detail or text.
  • Choose the aspect ratio for the final use. A 9:16 phone wallpaper and a 21:9 banner need different compositions.
  • Start with four images to explore a few directions before refining one.
  • When a result is close, change one thing in the prompt at a time. Then you know what made the difference.