Variation: see the neighbors before you commit

Variation generates two to four sibling images from the art on your canvas so you can compare close alternatives or build a series that visibly belongs together.

A single proof is a sample of one. It may be the best version of your idea or the third best, and looking at it alone tells you nothing about which. Variation settles that by producing siblings of the same art in one run — same idea, different light, framing, color and detail — so the decision becomes a comparison instead of a guess.

Only the art layer is generated. Your title and author name remain live vector type above it, identical across every sibling, which is exactly what lets you judge the art rather than the typography.

How to use it

1. Put the art you want siblings of on the canvas
   The scene takes its source from the canvas automatically. Run it on the proof you almost like: variations of a weak image are several weak images, so it is worth correcting the direction with Image to image first.
2. Set how far the siblings may stray
   The slider runs from close to far. Close holds the palette and framing near the original and varies detail, light and texture; far lets the model reconsider the scene itself. Series work usually lives at the close end and exploration at the far end.
3. Choose two, three or four
   The count chips decide how many images come back in a single run. Three is the useful default — two rarely reveals a pattern, and four earns its keep when you are exploring rather than choosing.
4. Run it and keep one
   Each image costs one credit, so a run of three costs three, and the button states the total before you press it. Results arrive as a grid: place the one you want on the canvas, and the others stay in this session's proofs in case you change your mind.

What it solves

Book three has to look like books one and two

Romance series live or die on shelf recognition. Take the art from the previous book, set the slider to close, and generate three siblings: the palette and framing carry over while the scene shifts. Pick the one that reads as the same series at thumbnail size, not the one that is prettiest at full size.

Choosing between two directions with real readers

Newsletter and ARC-group A/B tests only work when the options are comparable. Two siblings of the same art differ in exactly the ways you want tested — light and mood — instead of differing in everything at once, which keeps the vote about the cover rather than about two unrelated pictures.

A proof that is eighty percent right

You like it, something nags at you, and you cannot name it. Three close siblings usually expose the answer immediately: it was the eye line, or the height of the horizon, or the coldness of the light. That is faster and cheaper than rewriting the brief and hoping the model reads your mind.

Frequently asked questions

Will the variations keep the same composition?

No, and no setting promises it. Close keeps them recognizably related — similar palette, similar framing — but the model is generating new images rather than nudging elements around, so figures, props and horizons can move. A realistic expectation for a run of three is one or two you would genuinely use.

How many credits does a run cost?

One credit per image returned. Choosing four spends four on that run whether you place one of them or none, because every image is generated before you see it. That is why three is the usual compromise between a real comparison and a wasted spread.

Can I go back to a sibling I did not place?

Yes, within the same session. Every image you generate is kept in the session proof strip, so placing one does not destroy the others and you can swap later. Once you leave the studio, treat the rest as gone and make sure the one you actually want is the one on the canvas.

Related tools

All 11 AI tools →