Smart Edit (AI editing)

Smart Edit lets you change an artifact by describing what you want instead of doing it by hand. Tell the AI the change in plain language and it regenerates the artifact for you. It is the fastest way to iterate when you know the result you want but not the exact steps to get there.

Open Smart Edit

Smart Edit lives inside the editor for any artifact you've generated.

• Open an artifact from Artifacts (or generate one from chat).
• In the editor's top bar, select the MotherFluxer logo — the first tab on the left, ahead of the artifact's own tool menus.
• A menu's search is another way in: press Ask AI (the MotherFluxer mark in the search) with nothing typed, and the menu closes and Smart Edit opens on Discuss with an opening line about what you were looking for — your next message carries that search with it.
• Smart Edit opens on Discuss, where you talk an edit through with an AI that can see your artifact. Type the change you want into Discuss your edits with AI… and send it.
• When the AI knows what you're after it offers a Suggested edit card with an estimated cost and a Generate button. Select Generate to run it. If the edit still needs something from you — a painted region, a chosen clip — the button reads Set up & generate and takes you to those controls first. When the edit works on one part of your artifact — a layer, an item, a track — the AI can name that part from what your artifact holds, and it arrives already picked in those controls for you to confirm or change.
• Before the AI offers a suggestion, it checks it the way Generate will: that every part it names is in your artifact, and that an AI model can take what the edit would send. A suggestion that would be turned away is corrected in the conversation instead of reaching you; what can't be known yet — the area you will paint, for example — is checked when you generate.
• The AI regenerates the artifact and saves the result as a new version, so you can compare it with the one before and roll back.

Each Smart Edit costs credits. You'll see the cost before you run, and you can top up anytime with Add Credits.

The three Smart Edit screens

Three icons sit in a column on the left edge of Smart Edit, and each one opens its own screen. The icon for the screen that's open is highlighted; select it again to close that screen.

• Discuss (the speech bubble) is where every edit starts, and it's what Smart Edit opens on. Your conversation reads down to the Discuss your edits with AI… line at the bottom, and scrolling up to reread earlier messages scrolls that line out of view. That line works like the chat window's message box: the paperclip on its left attaches a file for the AI to look at — a photo to match, a document to read — and it stays in the conversation so you can refer back to it, and the microphone on its right dictates your message until you have typed something, when it becomes the send arrow. Closing Discuss leaves the controls for whatever edit is currently set up; if no edit is set up yet, you get No edit is set up yet — describe the change in Discuss to get started. and a Discuss the edit button that takes you back.
• Cost (the dollar sign) lists what your Smart Edits on this artifact have cost so far, one line per edit, named for the kind of edit and the model that made it, with the total at the bottom. Where an edit ran more than one charge, the arrow at the start of its line opens them; a run of Discuss messages shows as one Discussion line that opens the same way. Each edit's line carries a ☆ to save that edit's recipe to Favorites and thumbs to rate the result.
• Favorites (the star) holds the recipes you've saved for this artifact type — until you save one it reads Favorite AI recipes over No favorites yet. Save one with the ☆ on any Suggested edit card or on an edit's line in Cost; selecting a saved recipe later hands it back to Discuss as a starting point, and the AI adapts it to the artifact you have open rather than replaying the old request. The arrow beside a recipe opens it: a recipe always keeps the model it was saved with, you can adjust its settings before you use it, and the bin removes it. A recipe saved from a suggestion to make a new artifact starts a fresh one instead — describe what you want and select Create.

Smart Edit opens as a short panel under the top bar. For more room, select the expand arrows at the bottom of the icon column: Smart Edit then fills the screen down to the bottom bar, and Discuss gets the same message box as the chat window, fixed at the bottom with your conversation above it. Select the arrows again to shrink it back.

How Smart Edit works per artifact type

Every type starts the same way — you say what you want in Discuss — but what you get back differs.

Images, Audio, and Video hand you controls for the medium. The AI's suggestion arrives with the right controls attached: a region to paint on an image, a clip to pick on a video timeline, a track to pick in an audio piece. You finish that setup and generate. An image edit also asks what it works on, beside the Generate button: Image changes the picture alone and leaves everything placed over it on top, while Flatten bakes everything you see — the items over the picture, its colour and its crop — into the picture first and edits that.

3D and Apps produce a whole new version. You describe the change in plain language and the AI regenerates the whole artifact as a fresh version — for example, ask a 3D model to be restyled or an App to behave differently. The one exception is 3D's Change settings, which changes the model's own settings from your words rather than regenerating it.

Documents have their own set of AI tools. Beyond rewriting the whole document from a prompt, you get one-tap edits for selected text, AI reviews that leave your document untouched, web research that writes a cited document for you, and layout cleanup. Each has its own section on this page.

Give an edit your pictures

Some edits work from a picture of yours as well as your words — a style to copy, a face to swap in, several pictures to combine, the first frame of a new clip. Such an edit shows one picture row under its name, the same on every editor: the paperclip brings a picture in from your device, and Pick image chooses one you already have. Until you add one it reads No image picked, and each picture you add shows as a small tile with its own remove button. An edit that combines several pictures gives each one a line beside it, What to keep from this image, and stops taking more once it holds as many as it can use.

• Documents and Apps take pictures too. There the row sits under the edit's model choice, and the AI looks at what you add — a mockup to build, a screenshot of a bug, a layout to match.
• Pictures from Discuss come along. A picture you attached to your Discuss message goes into the edit the AI suggests, so "make it look like this" opens with your picture already in place. Whatever else the suggestion names, such as the track it works on, is already picked too.
• Each edit starts with an empty row. Switching to another edit, or running one, clears the row; the pictures themselves stay in your Artifacts.
• A picture still on its way holds the edit. Generate waits while a picture uploads, and an edit whose picture can't be made ready isn't sent, so it never runs without it.

Painting a region (Images)

Some image edits work on a specific area rather than the whole picture. When the edit the AI suggests needs a target, you'll be asked to paint the region first.

• Choose Set up & generate on the suggested edit.
• Mark the part of the image you want the AI to change. How you mark it is chosen in the Smart Edit panel's area row: paint it freehand, drag a rectangle, or drag a circle, with Clear beside them to start the area over. The area itself is drawn on the picture.
• Check the description and run it.

This tells the AI exactly where to work, so the rest of your image stays untouched. The area follows the picture wherever it sits on the page — moved, resized, turned or cropped — and any part of it past the picture's edge is left out. Two areas are turned away before the edit runs: one that misses the picture altogether (The area you marked is not on the picture. Mark an area on it and try again.), and, under Image, one on a picture you have tilted in perspective (Area edits do not work on a picture with perspective. Reset its perspective and try again.).

On a picture larger than the edit's AI model takes, only the area you marked, with a margin round it, is sent, and the rest of the picture stays exactly as it was. If the area you marked is itself larger than the model takes, your device scales it down to fit before it is sent and says so for a moment — The area you marked is larger than this model takes, so it was scaled down to fit. — so that area can come back a little softer than the rest of the picture.

Tip: Describe what should appear in the region — for example, "a wooden bench" or "a red bow tie" — rather than phrasing it as an instruction. The painted area already tells the AI where to work, so a short description of the thing you want gives the cleanest result.

Chain edits together (Images)

When you talk through an image change with the AI, it can suggest a sequence of edits that run one after another — for example, sharpen the photo and then restyle it. Each step builds on the result of the step before it, so you can get to a layered result without running and re-opening the edit yourself.

• The AI lays out the steps in order, numbered, with a short description of what each one does.
• Review the sequence, then select Generate once to run the whole chain.
• The edit works through the steps in order and shows you which step it's on (for example, "Step 2 of 3").
• Every step is saved as its own version, so you can compare the stages and roll back to any point in the chain.

If a step can't be completed, the chain stops there and any steps that already finished are kept as versions — so you never lose the work that succeeded. Chained edits are available for images.

Editing video (Video)

Video has six edits for working with what is on your video's timeline. Describe what you want in Discuss and the AI suggests the one that matches, then generates the result.

• Add clip generates a brand-new clip from your description and drops it onto the timeline. You can optionally add a reference image for the clip's first frame so the result resembles a look you already have; without one, the clip is generated from your words alone.
• Extend clip continues an existing clip seamlessly from its last frame, so the new footage picks up where the original left off. Choose which clip to continue with the Clip control, and the continuation is added right after it on the timeline.
• Edit clip regenerates or restyles an existing clip in place — it swaps the footage while keeping the clip's spot on the timeline. Choose which clip to change with the Clip control.
• Add sound listens to a silent clip and generates effects and ambience that fit it. Choose the clip with the Clip control and, optionally, describe the sound. If the AI can't take the clip's footage as it is, your device first makes a copy it can take, and the line under the edit says Reformatting video at model's request until it's done.
• Animate an item writes how items arrive, move and leave from your words, such as "bounce in, then float". Choose what it works on with the Item control, which starts on what you have selected; pick several and they go in turn. A text can move by letter, word or line, so say so when it matters: "letters drop in one at a time". For a movement no ready-made animation gives, such as a swing that dies away, it writes a curve on the item's position, turn, size or fade, and that setting's Source then reads Written by AI. The result replaces how each item arrives, moves and leaves, and any curve on the setting it wrote; a drawn path, a speed ramp and copies stay. It is an ordinary animation you can adjust in the item's Animate menu.
• Change settings tunes items already on your video from your words, such as "make it warmer and fade it out": an effect's settings, where an item stands, how see-through it is, or a curve that changes a setting while the item shows. Choose what it works on with the Item control; pick several to change them in one edit. It never adds an item, an effect or footage.

Two controls shape the edits that make a clip: a Model control, where you pick which model generates the clip, and a Length control, where you set how many seconds the new clip runs. The lengths you can choose from depend on the model you pick. Like any Smart Edit, the result is saved as a new version so you can compare and roll back.

Preview the look first (Video, optional)

You don't have to preview. Describe your clip and select Generate to produce it directly: Add clip builds from your words alone, while Extend clip and Edit clip continue from the clip you chose. Because a video costs more than an image, every edit that makes a clip also offers an optional cheap preview so you can lock the look before committing to a full clip.

1. Preview look — generate a quick, inexpensive preview image of the clip's first frame from your prompt. You'll see its cost before it runs. If it's not right, adjust your prompt and select Regenerate to try again — as many times as you like.
2. Use this image — when the preview looks right, confirm it. That image becomes the first frame the clip is built from.
3. Generate — produce the full video clip. You'll see its cost before you run, and the finished clip is saved as a new version.

The preview is there when you want to dial in the first frame cheaply before paying for the full clip — and you can always skip it and generate straight from your description.

Quick text edits (Documents)

Documents give you one-tap AI edits for the text you're working on, so you don't have to write out a full instruction every time.

Select an element in your document — a heading, a paragraph, a text box — and a small button with the MotherFluxer logo appears beside it. Choose it to open a short menu of quick edits:

• Improve rewrites the text to be clearer and stronger, and corrects any spelling and grammar mistakes.
• Shorten makes it more concise.
• Elaborate expands it with more detail.
• Change tone rewrites it in a tone you describe, like formal, friendly, or persuasive.

Most run with a single tap. Change tone asks you for the tone first. Each one edits only the text you selected. On a text box or other element the result is saved as a new version; in a document's flowing body text it lands in place as an ordinary edit you can undo.

When you're editing a single element, a Context option controls how much the AI looks at:

• Just this element edits only the text you selected — the most focused and cheapest option.
• With surrounding also shows the AI the other elements on the same page as read-only reference, so the edit stays consistent with the text around it.

Every quick edit uses credits, and what it cost shows in the Cost screen for that document.

Review a document with AI (Documents)

Documents have two AI review options that give you feedback without changing your document. They never alter your text or create a new version — they just tell you what could be better, and you decide what to do.

AI critique adds inline comments to the parts of your document that could be improved. Run it, and highlighted spots appear in your document. Tap a highlight to read the comment — what to improve and why. From the comment you can:

• Acknowledge to dismiss that comment once you've read it.
• Ask AI to address to hand the comment to the AI conversationally — it opens Discuss focused on that element, proposes an edit that resolves the comment, and you accept it (or refine it first) before it's applied. The comment clears once the edit is saved. Use this when you want to talk through the fix rather than apply a canned one.
• Apply fix when the AI offers a concrete change — this makes the suggested edit for you right away and saves it as a new version.

Use Dismiss all to clear every comment at once. A review stays with your document until you close it out or make an edit: close the document and its comments are still there when you open it again, and the review is still a message in Discuss, where you can close it out too. Closing it out in either place clears it from both. The comments never show up in exports or previews.

Assess gives you an overall review in a side panel: a short summary, scores for things like clarity and structure, and lists of strengths, weaknesses, and the most useful improvements to make. It's the quickest way to gauge where a draft stands before you dig into specific changes.

Both options use credits, like any Smart Edit, and what they cost shows in the Cost screen for that document.

Compare reviews from multiple AI models (Documents)

Multi-model review reviews your document with several different AI models at once and lays their feedback out side by side, so you can compare perspectives instead of relying on a single opinion. Like the other reviews, it never changes your document.

Ask for a multi-model review in Discuss and select Run review on the card that comes back. After a short wait, each model gets its own column in a comparison grid, and every column shows that model's take:

• a short overall summary,
• scores for things like clarity and structure,
• lists of strengths, weaknesses, and suggested improvements,
• and a count of the inline issues it found, which you can expand to read each one.

When a model offers a concrete fix for an issue, Apply fix makes that change for you and saves it as a new version. Because different models often notice different things, the grid is a fast way to spot the feedback they agree on — and the gaps only one of them caught.

Because it runs several models at once, Multi-model review uses more credits than a single-model review. You can see what each run cost in the document's cost history.

Research and write a document (Documents)

Research & write turns a topic into a finished, cited document. Instead of starting from a blank page, you give the AI a subject and it researches the web, then writes a structured document for you — with inline citations linking back to the sources it used.

1. Open Smart Edit on a document and ask Discuss to research and write about your subject.
2. Say what you want researched and written — for example, "the economic impact of remote work since 2020."
3. Select Generate on the Research & write card that comes back. Researching and writing takes longer than a normal edit, so you'll see progress while it works.

When it's done, your document is filled in as a new version: a multi-section piece on your topic, with numbered citation markers in the text that point to the real sources behind each claim. It's the fastest way to go from an idea to a sourced first draft you can then refine by hand or with the other Smart Edit tools.

Because it researches the web and writes a full document, Research & write is the most involved Smart Edit and uses more credits than a typical edit. You'll see the cost before you run.

Tidy up the layout (Documents)

Documents have a set of Layout tools under Smart Edit for changing how your content is arranged — from a light cleanup to a full redesign. Each saves the result as a new version, so you can compare it with the original and roll back, and each uses credits with the cost shown before you run.

• Auto-layout cleans up how the elements are arranged without moving any content. If your headings, paragraphs, images, and text boxes have drifted out of alignment or started to overlap, Auto-layout realigns them, evens out the spacing, keeps everything inside the page margins, and clears overlaps.
• Apply template reflows your existing content into a named layout. Pick a template — such as a formal report, a two-column article, a cover page with sections, a résumé, or a formal letter — and the AI re-arranges and re-styles what's already there to fit that structure. It doesn't write new content; it gives your existing content a polished, consistent layout.
• Make presentation turns your document into a slide-style deck — one section per slide, each with a short title and concise bullet points, plus a title slide up front. It's the fastest way to go from a written document to a presentation-style layout.
• Rearrange lets the AI propose a better reading order for your sections and elements — leading with the most important content and grouping related pieces together — while keeping all your content and styling exactly as it is. Unlike Auto-layout (which only fixes alignment and spacing), Rearrange changes the sequence.

Open Smart Edit and describe the layout change you want in Discuss — name the tool ("apply a report template", "make this a presentation") or just say what you're after, and generate the card that comes back.

Mark spots for AI to fill later (Documents)

When you're drafting a document and want to come back to a spot later, you can mark it as an AI placeholder instead of writing it now. Select the words you want to replace in a text element, open the floating menu beside the selection, and choose Mark for Followup. A small box asks what the AI should write there — for example "expand into a paragraph" or "add a one-line summary." The marked text gets a dashed violet highlight so you can find it again.

Placeholders stay with your document when you save and reopen it, and they never appear in exports or previews — they're just notes to yourself and the AI.

When you're ready, open Smart Edit and close Discuss with the speech-bubble icon — an AI placeholders list with everything you've marked sits behind it. Choose Resolve to have the AI fill them all in at once, matching your document's tone — it saves the result as a new version. To drop a placeholder you no longer want, use the ✕ next to it in the list; that leaves the original text untouched.

Resolving placeholders uses credits like any Smart Edit, and what it cost shows in the Cost screen for that document.

Choose the AI model (where available)

For many edits you can pick which AI model runs them, instead of always using the default. When an edit supports it, a Model control appears with the available models for that operation.

• Each model is labeled with a short quality/speed hint, and one is marked as the recommended choice — the best fit for that edit. If you don't pick, the recommended model is used.
• Different models can produce different results, take different amounts of time, and cost different amounts. The cost shown before you run reflects the model you've chosen, so you can weigh quality against price.
• Model selection is available across Image, Document, and App edits, with Audio and Video offering their own model and voice choices in their pickers. Where an edit has only one suitable model, no Model control appears — there's nothing to choose.

Picking a model is optional. The recommended default is there so you can just describe the change and run it; the control is for when you want to steer quality, speed, or cost yourself.

When an edit is too big for its AI models

Every AI model has limits on what it takes in — how large a picture, how long a clip or a recording, how long a description. An edit that is past those limits is turned away before it runs, so you aren't charged for it.

• When none of the models for that edit can take it, the message begins None of the AI models for this can take what you sent and, where it can, says how big what you sent is and the most they take. Try a smaller or shorter file, a shorter description, or different settings.
• When you picked the model yourself and it can't take it, the message says so by the model's name, with how big what you sent is and the most that model takes where it can, and lists the models that can take it, so you can pick one of those instead. When no other model can take it either, it asks you to send something smaller.

A painted area on a large picture is the exception: rather than turning the edit away, your device sends only the area you marked, with a margin round it, and scales that down to fit if it is still too large.

What the AI can look up while you talk

When you talk an edit through with the AI instead of running one straight away, it can go and check things before it suggests anything — so its suggestion is based on what is actually there rather than on a guess from the title.

• It can read the artifact you're discussing. For a document that means reading the text, including scanned and image-only PDFs. For an image it means looking at what is really in the picture.
• It can look things up on the web when a suggestion needs current or outside knowledge — a search, a specific page you paste a link to, a researched answer with sources, or a plain factual lookup.
• It only ever looks at the artifact you have open, and only ever at your own work. It cannot be talked into reading someone else's file.

You don't ask for any of this directly — the AI decides when a lookup would help. Each lookup uses a small amount of credits, and it shows up in the Cost screen for that artifact.

Pressing Generate twice

If you press Generate twice in quick succession — or a flaky connection makes your device retry on its own — only one edit runs, and you aren't charged twice. The second press comes back as "Already sent"; the first edit carries on and its result arrives on the artifact as usual.

Pressing Generate again deliberately, after the first edit has finished, is a different matter: that's a new edit, and it runs and is charged as one.

Stopping an edit that's running

Press Stop while an edit is running and it stops — on our side as well as yours. The AI is told to put its work down, your artifact is left exactly as it was, and nothing new is saved over it.

You're still charged for whatever the AI had already produced at the moment you pressed Stop, because that part genuinely ran. Stopping early therefore costs less than letting an edit finish, and often much less.

Stop applies to that one edit. A chat you have open, or an edit running on a different artifact, keeps going.

If you stop a chain of edits partway through, the steps still to come don't run and aren't charged. Steps that already finished are kept — each one is a version you can go back to.

If you press Stop just as the edit lands, it may already be finished — in that case the result appears as normal, and you can step back to the previous version from the artifact's history.

An edit isn't available for this artifact type yet

Not every kind of edit exists for every artifact type yet. Ask for one that doesn't and the AI says so in Discuss rather than running something else; on a phone, an edit that isn't live yet is listed with a Coming soon label you can see but not select.

There's nothing to fix on your end — that edit simply isn't ready. In the meantime, describe the result you want and let the AI suggest the closest edit it can run, or open the artifact's tool menus and make the change by hand.

What's the difference between Smart Edit and editing by hand?

MotherFluxer gives you two ways to change an artifact, and you can switch between them anytime. Neither is a mode you switch into — both are always one click away, on the same top bar.

• Smart Edit (the MotherFluxer logo, first on the left of the editor's top bar): describe a change in plain language and let the AI regenerate the artifact. Best for big or open-ended changes, or when you're not sure how to make the change yourself. Uses credits.
• The tool menus (the tabs that follow it along the same bar): hands-on, no AI. You make precise changes yourself with the editor's tools. Best when you know exactly what you want to adjust and want full control.

Use Smart Edit to explore and make broad changes fast, then switch to the tool menus for fine-tuning. Every result is saved as a version in the File menu, so you can move between versions freely.