# Ask the chat for a change > There are two editors, and two ways this works. The slides one applies the change right away, with an “Undo”; the text editor shows you a before/after to accept. Updated 2026-09-22 The assistant lives on the right of the editor, under the “Assistant” tab. Depending on the post format you are in the slides editor or in the text editor — and **they do not behave the same way**. Knowing which one you are looking at removes every surprise. ## In the slides editor: the assistant applies Carousel, image, story, pin: the post is a run of slides. There, the assistant **edits the document immediately**. It says so itself on the first screen: “Rewrite, add a slide, illustrate, change the tone: the assistant edits the document, and every reply can be undone in one click.” The field asks you to “Edit the post or the slide you have open…” and carries its estimated cost. While it works, every action shows up as a card in the thread, with its summary: “Slide 3 · title and text,” “Slide 5 · added,” “Caption updated.” Several actions of the same kind collapse into one card — “Slides edited,” “Image descriptions,” “Alt texts”… At the bottom of a finished reply, an “Undo” button puts the document back the way it was. If several things changed, it counts them: “Undo 3 changes.” Once undone, it reads “Changes undone.” > **Watch out** > > That button undoes **that one reply**. If you edited something in between, it goes disabled and points you to the history: “Later changes followed: use the history.” The toolbar keeps “Undo” and “Redo” for everything else. ## In the text editor: the assistant suggests Text post, thread, poll: no slides, a separate editor. There, the assistant **touches nothing without your say-so** — it is written at the top of the panel: “Ask for a rewrite: the assistant suggests, you approve.” You type your request — “Shorter, warmer, a question at the end…” — and the reply shows up as a **before/after**, word by word: what goes away is struck through, what arrives is underlined, post by post if it is a thread. Two buttons: “Apply” (`⌘` `↵`) and “Reject.” “Apply” takes the whole suggestion at once. Only then does the card read “Applied,” and an “Undo” appears in its turn. > **Tip** > > To rework just one post of a thread, skip the assistant: each post’s AI actions menu — Shorten, Rephrase, Punchier… — offers the same before/after mechanism, with “Accept” and “Reject,” on that post alone. ## The open slide is context, not a scope In the slides editor there is **no “scope” to set anymore**: no pill, no button to target a slide. The slide you are looking at serves as context. - A request that clearly names one slide applies to that slide. - A request about the post — “shorten all the texts by 30%” — applies to the post. - And when the assistant cannot tell which it is, **it asks** instead of guessing. ## Images and references The assistant can illustrate: it writes an image’s description — the “Image descriptions” cards in the thread — then generates it. When a person, product or place from the project has to appear in it, the assistant calls it by its @handle instead of describing it: it knows the references in “In this post” and those in the library. Three at most per image, and a handle that matches no reference is refused. A reference it calls joins “In this post” on its own — see [call a reference with @](/en/guide/call-a-reference-in-an-image/). You can also call a reference yourself: type `@` in the chat field, and the picker offers the ones in “In this post”, then the rest of the library. Your pick becomes a chip, with its thumbnail and name. Two references that look alike (Bastien and Bastien B.) no longer get mixed up, and a library reference that isn’t in the post yet counts as an explicit request. In the thread, references read as chips, in your messages and in the assistant’s replies. `↵` sends, `⇧` `↵` adds a line. ## Retouch an image Ask for a change **inside** the existing image — “remove the cup on the counter,” “warmer light on slide 2” — and the assistant suggests a **retouch** instead of rewriting the description. A card is added to its reply: “Retouch the image · Slide 2,” the thumbnail, the instruction in quotes and the estimated cost. - That card **always** waits for your go-ahead, whatever it costs: “Retouch” starts it, “No thanks” sets it aside — “Retouch skipped.” - While it runs, “Retouching · 12s” and a “Stop” button. When it’s done: “Slide 2 · retouched, v4,” with the actual cost. - It’s a **new version**: the old one stays in the image’s versions, and the reply’s “Undo” button takes you back, retouch included. - If it fails, the card says so — “The retouch did not go through. Nothing changed in the image.” — and offers “Retry.” For a different picture — another subject, another scene — the assistant rewrites the description instead, and generating it goes through the approval card below. The same rule applies in an animation’s “Assistant” tab, for a moment’s background photo. And you can retouch by hand too, from the “Image” panel: [image versions, comparing and retouching](/en/guide/image-versions-compare-and-retouch/). ## When the assistant asks for your go-ahead A rewrite goes out without ceremony. Generating images does not: as soon as several are needed, or a single one costs more than ✦ 5 — which any image in high quality does, and high is the default for posts — a card slides into the thread — “Generate 6 images,” with its cost — and waits for “Generate” or “Dismiss.” Once decided, it collapses to a single line: “Started” or “Dismissed.” While it works, a dot appears on the “Assistant” tab: you can switch to “Slide” or “Post” without losing the thread. “Stop” interrupts the reply in progress, “New conversation” starts over, and every reply can be copied. ## What’s next Whatever the assistant does, it does through **operations**: visible, explained, undoable. That is the subject of [accept, reject or undo an operation](/en/guide/accept-reject-or-undo-an-operation/). And for pictures, [generate a consistent image series](/en/guide/generate-a-consistent-image-series/). --- Source: https://joli.studio/en/guide/ask-the-chat-for-a-change/