Point your agent at a picture and it works from what it sees: a style to match, a product to keep, or a result to edit. There are two ways to add one.
Both guide the very next prompt — the only difference is where the image comes from.
Click the Click or drag an image here tile beside the prompt box, or drag files in. JPEG, PNG, WebP, BMP and AVIF work, several at once. Thumbnails line up in a strip where you can drag to reorder, remove one, or clear them all.
Want to keep a reference but skip it for one prompt? Switch it off — it stays in the strip, unused, until you switch it back on.
Hover any generated image — in the chat or in Creations — and tap the circle in its corner. A checkmark means it’s selected, and it moves into the prompt box as the input for your next message. Select more than one to combine references.
This is how you refine: generate an image, select it, then describe the change. Each round builds on the last instead of starting over.