Create and edit an image

Brief an image, add reference photos, choose quality and model, review the takes, and change one part while keeping the original.

Updated 2026-10-22

Make your first image

1. Open New chat and describe the image: the subject, setting, light and style, where it will be used and its shape, such as a square product image or a vertical story.
2. If the image must show your product, logo or a person you have photos of, select + beside the brief and choose Upload files, or choose Add from Library to reuse an item you already have. Say in the brief what each file is for.
3. If the image needs words on it, quote them exactly and say where they go.
4. Open the quality menu beside Send if you want to change the quality level or pick an image model. Leave Images on its default to let the agent choose.
5. If cost matters, open Planning budget in the same menu and choose Check what this message costs, then press Send.

Make a square product image of our ceramic pour-over kettle (photo attached) on a sunlit oak counter, morning light, shallow depth of field. Keep the kettle's shape, color and logo exactly as in the photo. No other words or logos in the picture.

Check that it worked

1. The result appears in the conversation. When the agent made several takes of the same image, each is labeled Take N of M.
2. Select Enlarge to see the image larger, or Open original in its More options menu to open the original file in a new tab. Check faces, hands, product details, edges and every word.
3. If something is wrong, say what to change in your next message. See Change one thing in an image.

An image existing in the conversation does not mean it was delivered. When the agent finishes, its reply says what it delivered and what is still missing.

What to send

A clear brief does more than a long one. Include:

• the subject and what it is doing;
• the setting, light and style, for example "studio photograph" or "flat illustration";
• where the image will be used and its shape, for example "Instagram story, 9:16" or "728x90 banner";
• any words that must appear, quoted exactly, and where;
• what must stay exactly as in your references, and what the agent may invent.

Say what people wear and what is in their hands. Describe food as a stylist would. Words like "sparkle" or "glossy" tend to appear literally, often more than you meant.

Reference photos

References tell the agent what something really looks like. Use them for your product, packaging, logo, a person or a place you want shown again.

• Give each reference a job. A product photo defines the product; a style reference defines the look. They are different constraints.
• For a person, product or place that appears in several images, the agent usually makes reference stills first and builds each image from them.
• A reference does not guarantee an exact match. Check labels, small text and logos at full size: image models can redraw them.

To reuse an earlier result as a reference, choose Use in chat in its More options menu. Your brand kit's logo and colors are another source. See Brand kits and Use references, characters and styles.

Choose quality and model

Open the quality menu beside Send. It shows Balanced by default.

If you or your workspace saved other default models, images start on those instead.

To pick the image model yourself, choose Show model choices and set Images. A model you pick that differs from your saved default is a pin: the agent makes new images for that message only with it and never swaps it for another. Crops, resizes, upscales and similar edits are not held to the pin. If the pinned model cannot do what the brief needs, the agent stops and tells you instead of switching. Adult images are the one exception; see Adult (18+) images. Leave Images on Auto to let the agent choose per image; it may use another model when the brief needs it, for example a model that sets type better. Model choices apply to your next message.

The models you can pick depend on your workspace. They can include partner models such as Gemini 3 Pro Image, Gemini 3.1 Flash Image, GPT Image, Grok Imagine Image, Seedream and Ideogram, and Wavemaker's self-hosted models. See Models and quality.

Shapes and sizes

Say the shape you need in the brief. Gemini image models, the default, make these aspect ratios: 21:9, 16:9, 4:3, 3:2, 5:4, 1:1, 4:5, 2:3, 3:4, 9:16, 4:1, 1:4, 8:1 and 1:8.

For an exact pixel size, such as an ad unit, the agent fits the finished image to the canvas. A still is never stretched: it is scaled by one factor and cropped to fill the canvas. A still fitted to a standard display ad unit comes back as a JPEG under that unit's file size limit.

For a Story or Reel, the agent usually makes a square image with all its words first, then extends it to 9:16 with an image edit that continues the scene above and below, so no words sit under the app's buttons.

How many takes

A take is one generated version of the same image. One request can ask for up to 8 takes. The agent decides how many to make from your brief and quality level, unless you say. For an ad image, it usually starts two takes at once, checks both and delivers the first that passes.

To see the takes of a result, open its More options menu and choose See the other takes. Every take is charged, including the ones you don't keep.

How review works

Before it calls an image done, the agent can check it with Wavemaker's review tools:

• the words in the image are read back and compared with the lines you asked for;
• the positions of words and logos are measured against the format's safe margins;
• a vision model looks at people, clothes, hands, proportions and seams.

A take with defects is fixed with an image edit or replaced by a new take. Review calls are model calls and are charged to the message like any other step.

The review is advisory. It does not replace your own look at full size, especially for small text, labels and faces.

Change one thing in an image

Name the image and the one change you want, and say what must stay the same.

Keep this kettle, the counter and the light exactly as they are. Replace only the window in the background with a plain warm cream wall.

The result's More options menu gives you three ways to aim a change:

• Refine opens Refine this output. Choose Whole output or Part of the output, then Use as target. Your draft shows Refining the whole output. For a part, name it in your message, for example "the kettle's handle". Nothing is sent until you write the change and press Send.
• Make a variation puts the image in your draft as the target for a variation.
• Paint a mask marks the area that may change. See Paint a mask.

Each change is new work with its own charge. The original stays in the conversation. When a newer version exists, the menu offers Compare with version N, which opens a slider between the two versions. See Choose, compare and revise a result.

Paint a mask

1. Open the image's More options menu and choose Paint a mask.
2. Paint over the part of the image to change with Brush. The status line shows how much of the image the mask covers. Use Erase, Undo, Redo and Clear to correct it, and Brush size to change the brush.
3. Optionally give the mask a name in Mask name (optional).
4. Choose Save mask. A notice confirms the mask was saved to this conversation's assets.
5. In your next message, describe the change for the masked area.

Keyboard: B for the brush, E for the eraser, [ and ] to change the size, and Ctrl+Z or Cmd+Z to undo (add Shift to redo).

The mask is saved at the image's own pixel size to the conversation's assets, where the agent finds it. Saving a mask does not change the image or spend credits. The mask tells the agent which area may change. Mask the whole object you want changed: a part left outside the mask can survive the edit. Check the rest of the image afterward, since an edit model can still redraw details. Paint a mask is offered only inside the conversation that made the image, and only when you can edit that conversation.

Other edits

Ask for these in the conversation. Each makes a new file and leaves the original unchanged.

For a logo, upload a transparent PNG or SVG if you have one.

What affects the cost

• The model. At the same size, Gemini 3 Pro Image costs more than Gemini 3.1 Flash Image, which the Draft level starts on.
• The size and references. Partner image models are priced by output size; larger sizes can cost more. Reference images sent to Gemini or Grok Imagine add to the call's price.
• The number of takes. Every take is charged, kept or not.
• Edits and fixes. Each edit, upscale or new take is its own charge.
• Review and the agent. The agent's own model use and its review calls are part of the message's cost.

Choose Check what this message costs under Planning budget before you send. It answers with an estimate, about one amount and up to a higher one. Checking spends nothing and does not limit the message. To set a ceiling, enter it in Total planning limit (credits) in the same section: sending reserves that amount, the reply stops when it is spent, and unused credits come back. See Understand credits, reservations and charges.

If something goes wrong

Work that finished stays in the conversation if a later step fails. Sending the same brief again is new work and is charged again. See Recover interrupted work.

Adult (18+) images

Wavemaker enforces these rules on adult images:

• adult stills are made on Wavemaker's self-hosted Krea 2 model, whatever image model you picked, and the agent tells you when it does this;
• adult work may not use a photo or video that shows a real person, and may not depict a real person; such a request is refused;
• sexual content involving minors is always refused;
• an image that is not rated 18+ is checked before delivery, and one that shows nudity or sexual content is not delivered.

Ordinary work uses real photos, logos, products and people freely.

Download and use

Use Download on the result (Download image on a result card) to save the original file, or Open original to open it in a new tab. Confirm the dimensions and file type for your destination before you use it. Wavemaker adds no watermark or badge to images. Making or downloading an image does not publish it. See Download results and Share and export.

Related guidance

• Brand kits and watermarks
• Understand credits, reservations and charges
• Choose quality, models and presets
• Use references, characters and styles
• Choose, compare and revise a result
• Video, speech, music and finishing