Default
Pick by where it will be posted. 4:5 fills the most feed space on Instagram/Facebook and tends to get the most engagement; 9:16 is required for TikTok, Reels and Stories; 1:1 is the safe all-rounder. Leave the prompt empty to simply reframe your pictures to this shape β no AI, nothing invented, exact pixels. With a prompt, the model generates instead and lands only close to the ratio.
That reads as a question. Image models follow orders, not requests β they often reply with a near-identical picture. Try "make the background dense jungle foliage" rather than "can you make the background full of nature".
Paste an image with Ctrl+V, or drop files here
The AI designs the poster, then your own product picture is laid over the one it drew β so the label, the flavour name and the small print are your real pixels, not the model's guess at them. Untick only if you want the AI's version of the pack.
On by default: after each run the result replaces your source picture, so the next instruction builds on it and anything you already fixed stays fixed. The thumbnail is labelled with the round number. Adding a new picture starts a fresh chain. The header/footer band is never fed back β it is re-applied to each result, so it cannot stack up.
β© start over from my original upload
Optional, but it roughly doubles the hit rate: the model is told the exact characters to look for instead of working out which lettering is the address. If your logo shares letters with the domain, naming the exact text also stops the logo being taken away with it. Say which corner if you know; a wrong guess sends the model to delete whatever is in that corner instead.
You asked for that in the prompt β tick the box above instead. A typed request is weighed against the "preserve captions and brand marks" rule and usually loses; the box overrides that rule outright.
Use this rather than typing "remove website name". The instruction sent with every edit tells the model to preserve captions and brand marks β which describes your address too, so a typed request fights it and the address survives, or comes back on the next run. This box states the removal as an explicit exception to that rule, and has the model rebuild what was behind the letters instead of patching over them. Works on its own: tick it, leave the prompt empty, and the address is the only thing touched.
Your products are placed on the new background automatically.
Adjust placement
On for a product on a plain white or solid backdrop, which is most product shots. Turn it off if your picture is already a full poster or scene β the cut can spread into the artwork. It abandons itself if it starts eating the picture.
The AI never sees this, and it is never resized β it is drawn at its exact pixel size, so your logo stays byte-for-byte as you made it. If it is wider than the generated picture, the picture is scaled up to match instead.
Generated images will appear here.
Drag a box β then keep just that box (crop), or wipe what is inside it (erase)
No selection yet β drag on the picture.