# Image versions, comparing and retouching

> The “Image” panel reads in the order you work: the versions, what the image shows, the recipe, then generate or retouch. Every version keeps its recipe, and two versions can be compared on the canvas.

Updated 2026-10-02

An image often takes two or three tries before it is the right one. The “Image” panel is built for that: nothing gets lost, every attempt keeps what produced it, and you can tell at a glance what changed from one version to the next.

## Where to find it

In the slides editor, “Slide” tab, below “Layout” and “Content”: the “Image” panel, as soon as the slide has room for an image. In an animation post, the same panel is called “Background photo,” in the moment’s inspector. In a text post, it opens in a window, from the image’s thumbnail — see [carousel, image, text, thread, poll](/en/guide/carousel-image-text-thread-poll/).

On the canvas, a click on the image area **selects** it: a ring goes around the image, and its toolbar replaces the slide’s. “Leave the image” or `Esc` brings the slide’s toolbar back.

## The panel, in the order you work

The panel reads top to bottom, the way you work: what you have, what you want, with what, then generate.

- **“Versions”** — the images you have already tried, as thumbnails numbered v1, v2, v3 in the shape of the area. The version in use is circled; click another one to bring it back. Hover over each to see its recipe and its actual cost.
- **“What the image shows”** — the description: subject, gesture, light, framing. The field opens on the image’s style, “same as the post” or “its own”; “Change” or `#` picks another one — see [create an image style](/en/guide/create-an-image-style/). Never text to write inside the image; the slide lays its own on top, and a product style writes the few words of its components itself. Without a description, the AI starts from the slide text. And `@` calls a [reference](/en/guide/call-a-reference-in-an-image/) from here.
- **“Recipe”** — everything that goes to the model with the description: “References,” “From style” (what the style attaches: style sheet, product screenshots, components written from the text), “Series” and “Format” (the area’s). At the top, the number of attached images and what they add.
- **The generate button** — “Generate image” if there is none, “New version” otherwise, with its cost. The quality sits next to it; you can change it here as well as in “Post images,” and the app reminds you: “Quality applies to the whole post.”
- **“Finishing”** — the focal point, “Fill” or “Fit,” the “Overlay” under text laid over a background image, and the “Alt text,” which “Write” drafts for you.

> **The cost, ingredient by ingredient**
>
> Hover over the generate button to see the breakdown: the image at the size of its area and at the post’s quality, then the attached images — the references’ sheets, the series anchor — and the “Estimated” total. It is rounded only once, on the total. The actual cost then shows on the version.

## Every version keeps its recipe

“All versions” unfolds the full list, newest first. Each row tells you:

- **where the image comes from** — “Generated,” “Retouched,” “Uploaded,” “Media library” or “Reference photo”;
- **what it was made with** — the quality, the style, the attached references, “series attached”;
- **what changed since the previous one** — “description edited,” “with the series,” “high quality,” or a retouch’s instruction;
- **what it cost**, in actual Sparks.

Three moves per version: “Use” puts it back in place; “Reuse the recipe” brings its description back into the field, references included, without touching the image; and the trash can deletes it, except the one in use. The post keeps the last eight versions of each image.

> **Tip**
>
> A version that works can become a reference: its “Create a reference from this image” button opens the creation window with the image already in place. See [create a reference](/en/guide/create-a-reference-from-photos/).

## Compare two versions

“Compare,” under the thumbnails, or the two-column icon in the image toolbar, puts **two versions side by side** on the canvas, each in the real slide, with its recipe underneath. The “Left version” and “Right version” menus change what you compare; “Use v3” keeps the good one, and “Leave” or `Esc` takes you back to the slide.

## Retouch instead of starting over

An image that is almost right doesn’t deserve a whole new generation: it deserves a **retouch**. The “Retouch” button sits under the generate button, and in the image toolbar on the canvas. You can also ask the assistant for it, in plain words: it suggests the retouch in a card that waits for your go-ahead — see [ask the chat for a change](/en/guide/ask-the-chat-for-a-change/).

1. **Say what should change**

   “What should change, and only that…”: a short instruction, or one of the suggested ones — “More light,” “Frame tighter,” “Change the background,” “Remove a stray object.”

2. **Launch it**

   The “Retouch” button carries its cost: that of an image at the size of the area, plus the retouched image, which goes out with the instruction. `⌘` `↵` launches it too.

3. **Keep the better one**

   “Retouching creates a new version; v2 stays in the versions.” The new one arrives marked “Retouched,” with its instruction in the recipe. If you like it less, one click on v2 brings it back.

## On the canvas

Once the image is selected, its toolbar does the rest without leaving the slide:

- the arrows flip through the versions **in place** — “v2 / 3” — so you judge them in the real layout;
- “Crop” lets you drag the focal point right on the image, and the arrow keys nudge it;
- “Replace” opens “Import a file…” or “Media library…,” free: the imported image becomes a version like any other;
- “Retouch” and “New version,” with their cost.

## See what goes to the model

Under the generate button, the folded “See what goes to the model” line shows the prompt as the model receives it: read-only, in English, each part colored by where it comes from — “Description,” “References (from their profile),” “Components (written from the slide),” “Image style,” “Framing of the zone,” “Fixed rules,” “Series,” “Style sheet,” “Product screens” — then the attached images, numbered as in the prompt. “Copy” puts it on the clipboard. It is there for the curious, and for writing to us when an image surprises you.

## What’s next

Making every image of a post hold together is the series’ job: [generate a consistent image series](/en/guide/generate-a-consistent-image-series/). And for the same person to come back from one image to the next, [call their reference with @](/en/guide/call-a-reference-in-an-image/).

---

Source: https://joli.studio/en/guide/image-versions-compare-and-retouch/