Image to image: change the art without starting over

Image to image rewrites the art already on your canvas along a direction you describe, and a faithful-to-free slider decides how far it is allowed to move away from the original.

There is a gap between a proof that is wrong and a proof that is nearly right. Nearly right is the expensive case: the scene works, the mood is off by fifteen percent, and a fresh brief would throw away everything that already worked. Image to image takes the art on your canvas as its starting point and asks for the same picture pushed in a stated direction — colder, emptier, later in the day, further from the subject.

The whole operation stays on the art layer. Your title and author name are live vector type on a layer above it, so they cannot be repainted, warped or misspelled however far you push the slider.

How to use it

1. Start from the art on your canvas
   Open the Image to image scene. Whatever is on the canvas loads into the INPUT column automatically — art from a brief, a file you imported, or the result of an earlier operation. There is nothing to upload.
2. Describe the change, not the whole cover
   The line answers one question: how should it change? Write the difference — colder light, empty the foreground, push the figure further away, make it winter. Re-describing the entire scene competes with the source image and gives a worse result than one clear instruction.
3. Choose how far it may travel
   The slider runs from faithful to free and lands in one of three bands: a subtle refinement that stays very close to the source, a balanced reinterpretation, or a bold one that takes real creative freedom. Start in the middle, move toward faithful when the composition is already right, and toward free when only the idea is worth keeping.
4. Run it, compare, then place it
   One run costs one credit and returns a proof with Before and After tabs so you can flip between them at full size. Place the version you keep on the canvas — and if a single small area is still wrong, repair that area instead of running the whole operation again.

What it solves

Space-opera art that reads as fantasy

Science fiction gets miscategorized when the palette is warm and the edges are soft: readers scanning the shelf file it under fantasy and scroll past. Ask for colder light, harder edges and a metallic palette at the balanced setting. The ship, the planet and the framing survive; the genre signal changes.

A cover that dies at thumbnail size

Amazon shows your cover roughly at postage-stamp size. If the art is a beautiful mid-tone scene with six competing elements, none of them survives that reduction. Run it near the faithful end with one instruction — raise the contrast and simplify the background — then check the result small before you place it.

A backlist title that looks like the year it launched

Design conventions move. A 2019 cover with a gradient sky and a small centered figure can be walked forward without commissioning new art: keep the subject, ask for the current look of the category, and run it at two or three slider positions to see how far you actually want to go.

Frequently asked questions

What does the slider actually change?

It sets how much of the source survives. At the faithful end the model is told to refine subtly and stay very close to the source image; in the middle it is told to reinterpret it in a balanced way; at the free end it is told to take creative freedom. That is one instruction, not a percentage — two runs at the same setting will not be identical.

Will the composition stay exactly the same?

No. Nothing here guarantees a pixel-identical layout, and even at the faithful end small things drift: hands, faces, the outline of a distant building, the number of windows. That is the honest trade for reinterpretation. When one area is wrong and the rest must not move at all, Region fix is the correct tool — it repaints a masked area and leaves every other pixel untouched.

Can I run it on art that was not generated here?

Yes. The source is simply whatever sits on the canvas, so imported photos, licensed stock and older cover files all work. The one thing the operation never sees is your text: titles and author names live on a vector layer above the art, which is why they stay razor-sharp in print while the picture underneath is rewritten.

Related tools

All 11 AI tools →