Generating images
The four canvas modes — Generate, Edit, Compose, Variations — and when to use each one.
The prompt bar at the bottom of every canvas has a mode strip with four pills: Generate, Edit (mask), Compose, and Variations. Each one is a different way to make a new generation, and each one lands as its own entry in the lineage rail.
Generate
This is the default mode and the one you'll use most often. Type a prompt, pick the controls, and click Generate.
Controls:
- Prompt — A natural-language description of the image. The more specific you are about subject, mood, lighting, and composition, the better the result.
- Aspect ratio —
1:1,16:9,9:16,4:5, or3:2. Pick the shape that matches where the image will end up (square for social, 16:9 for slides, 9:16 for stories). - Quality —
Low,Medium,High, orAuto. Higher quality costs more — the estimated cost shows next to the button. - Variations (n) — Generate 1–4 images at once for the same prompt. Use this when you're exploring directions and want to compare side by side.
Each canvas has a default aspect ratio you set when creating it. The prompt bar starts at that ratio, but you can change it for any individual generation.
Edit (mask)
Edit lets you change a specific region of the current image while leaving everything else alone. Useful when most of the image is right but one part needs work — swapping a background, fixing a hand, replacing text on a sign.
Open Edit mode
Click Edit (mask) on the mode strip. A mask editor opens over the current image.
Paint the area to change
Use the brush to paint over the part of the image you want the AI to modify. Anything outside the mask stays as-is.
Describe the change
Write a short prompt for what should appear in the masked area. ("Replace the sky with sunset clouds", "Make the shirt navy blue".)
Generate
Click Generate edit. The new image appears in the display, with the unmasked region preserved.
Compose
Compose blends multiple images into one — for example, putting your logo onto a generated background, or combining two reference shots. You pick 2–4 source images and write a prompt describing how they should come together.
Switch to Compose
Click Compose on the mode strip. A reference picker appears.
Add references
Pick 2–4 images. They can come from the current canvas's lineage, an upload, or other generations in the workspace.
Describe the composition
Write a prompt that explains how the references relate. ("Place the logo from image 1 onto the building in image 2", "Combine the lighting of image 1 with the subject of image 3".)
Generate
Click Compose. The result lands in the lineage as a new generation.
Variations
Variations make a new image with the same prompt and similar feel as the current one — slightly different framing, slightly different details. Use it when the current generation is close to what you want but you'd like to see a few alternatives.
You don't write a new prompt for variations. Just pick a count (1–4) and click Make variations. Each one becomes its own entry in the lineage.
Cost and quotas
The prompt bar shows an estimated cost in cents next to the generate button. The estimate is based on:
- The image size implied by your aspect ratio
- The quality tier
- The number of variations
Your workspace may have a monthly image budget. If you're close to the limit, the quota readout in the prompt bar will warn you before you start a generation that would exceed it.
Tips
- Be concrete about style. "A photorealistic close-up of…" or "A flat illustration in pastel tones of…" leads to more consistent results than "a nice picture of…".
- Iterate on the lineage. When a generation is 80% right, edit the masked region instead of restarting the prompt — it preserves the parts that worked.
- Lock down brand. If you have a brand style attached to the canvas, you can keep prompts simpler — the style does the heavy lifting on palette and feel.
- Use Variations sparingly. It's great for picking a final image, but if the current generation is fundamentally wrong, a new prompt is faster than rolling for variations.