Making images
Making and changing images with the Assistant, what happens while they are made, and opening an image in its editor.
The Assistant’s prompt floats at the bottom of a Canvas (press / to start typing in it). Hover it and a small handle appears: above it while it floats, just to its left while it’s docked. Drag the handle to the right edge, or press it and choose Dock right, to keep the whole conversation beside the Canvas; Float puts it back. It remembers where you left it. To add images from your computer, use Add images or videos (+) in the toolbar. Choose Image from the mode button at the start of the prompt (see Talking with the Assistant; in Ask the Assistant talks it through with you first), describe an image and press Enter: the prompt clears, ready for your next one, and the image’s place appears in the middle of your view straight away, glowing softly, like light behind glass, while it is generated, and the picture fades in when it is ready, usually within a minute. Until it is, the send button is Stop, which cancels it (and anything else you started that is still running) at no cost, and takes its place off the Canvas (see Stop). Meanwhile it is an ordinary image: move, resize or delete it, and the result lands where you put it. If you reload while it is being made, it carries on and still fills its place. If the Assistant can’t take a prompt at all, the prompt comes back with why (or, if you’ve already started the next one, why is shown without touching it). If it can’t be made, it stays in place, its glow turned red and still: hover it for Retry, or select it to see why in its properties panel at the top right, with Retry and Remove (delete it as any node). To work from an image, right-click it and choose Use as reference (or choose it in its properties at the top right); select several first to choose them all at once. Or drag it onto the prompt: the prompt lights up when it can take what you’re dragging, and the image stays where it was on the Canvas. You can also paste an image (a screenshot, say) into the prompt with Ctrl/⌘ + V: it becomes a reference without being added to your assets or the Canvas, and pasting the same image again adds nothing. If the model you’ve chosen can’t take that many images, the prompt says so and nothing is added. Select one image on the Canvas and the Composer shows it above the prompt, outlined in the selection’s colour: in Image, Editing selected image, so what you write changes it, and the result goes beside it; in Ask, Asking about the selected image, which the Assistant sees without anything being made; in Video, Selected image, the video’s first frame, which its menu can make the last frame, a reference, or Don’t use. It is never numbered, and it’s never one of your references. × on it deselects the image, and selecting something else changes it, so what is sent is always on screen before you send. With several images selected, nothing is chosen for you: pick one there to edit, or choose Use as references. With one image, the Assistant changes it: the changed image is a new image beside the original, which is left as it was. If there’s no room beside it in your view (the original is at the edge, or off screen), the new image lands in the middle of your view instead: what you make always arrives where you can see it. With several, it makes a new image from them; refer to them in your prompt by the names shown above it (Image 1, Image 2). Each one is shown above the prompt as it will be sent: remove any you don’t want with ×, which leaves the Canvas as it is. Sending takes them with the prompt, and once it’s accepted they’re let go (if it’s refused, they stay); choose them again for the next one. Any you choose while it’s being sent stay for the next.
Settings
The sliders button beside the cost meter opens Settings: the Model, then the settings that model offers. For an image, its shape (Aspect ratio) and Size, and for a model that has them, its Quality; for a video, also its Resolution, Duration and Sound. Hover the button (or listen to it with a screen reader) to hear what is set, such as Seedance 2.0 Fast · 16:9 · 5s. The cost meter beside the send button follows what you set. Settings are remembered in this browser: the model you chose for Image and for Video, and what you set for each model. A model you haven’t used before starts from its own recommended settings. Auto shape takes the shape of what you make it from (an image you edit or animate, or the first image you chose), as near as the model makes it, and never crops it; a shape you choose stays, whatever images you add or take away. In an image’s editor a change keeps the image’s shape and resolution, so only the model can be chosen there, and its Quality where it has one.
Quality is offered only by models that have quality tiers, such as GPT Image 2.6 (Low, Medium, High, XHigh and Max, starting at High; Medium costs less). A higher tier takes longer and costs more, sometimes many times more: the cost meter beside the send button follows the tier you choose, so check it before you send. Each image you send with it (the one you change, and each reference) adds a little too.
The Assistant picks a model for what you ask: Settings show Auto, and the button names the model Auto is using. You can choose another in Settings, which lists the models that can do what you’re asking, or name one in the prompt itself with --model, for example a red chair on grass --model flux-klein. The --model part is taken out before the prompt is sent, and Settings show the model you named; a name that isn’t a model’s, or a model that can’t do what you ask, is refused with why. Each model’s page lists its name (see Models). FLUX.1 Schnell (--model explore) makes Explore’s quick drafts that your plan includes each day (see Explore). Its cost is shown beside the send button before you send, as the cost meter (see Usage and prices). What the Assistant makes is an image like any other: select, move, resize, copy or delete it, and undo placing it. Every image it makes or changes is kept in your Space, with the prompt and what it was made from.
A style from another image
To make one image in the style of another (a photo as a woodblock print, a sketch in a painting’s colours), attach both and choose Image. Above the Composer, the images are laid out by what each is for: drag the image whose style to take into Style from (or choose Style from its menu). That image says Style under it. Roles, beside the images, opens this again once it’s folded away. What you send keeps the other image’s subject and composition and takes only the style image’s colours, line, texture and medium; anything you write is added to it. In an image’s editor, the image you’re editing is the one restyled, from the image you attach. You can also just ask the Assistant (“make Image 1 look like Image 2”). Style from is offered by models that do this faithfully: today Nano Banana 2.1, which Auto uses when you attach images.
Creating an image on the Canvas
Double-click empty space on a Canvas for a menu of what to make there (see Adding to a Canvas): Upload asset… places images or videos from your computer where you clicked, and Create image opens the image editor on a blank sheet, where the new image will go. Choose its shape at the top right (1:1, 4:3, 3:4, 16:9 or 9:16), then either describe the whole image in the prompt and press Enter, or draw Regions to lay it out: drag a rectangle where each thing goes and say what goes there (“a red bicycle”, “the title in white”), with words for the whole scene if you like. One request makes them all, each in its place. Regions guide where things go; a thing can reach a little past its rectangle.
Nothing is saved until the image is made: close the editor first (✕ or Esc) and nothing is left behind. Once it’s made, the image is on the Canvas where you double-clicked, and the editor opens on it, ready to change.
Editing an image
Double-click an image on a Canvas, or right-click it and choose Edit, to open it in the image editor: change it with words or Regions, work with its layers, masks and tools, relight, crop, extend, upscale or animate it. See The image editor.