Point at it. Say it. Only that changes.
One screen for every photo, video, page and email. The piece you're making sits big, looking exactly how your customers will see it. You point and talk. Olivia lights up what you mean, changes only that, and keeps the rest of the campaign in line.
- 1The piece you're working on. Big, and shown where it will live. A post looks like a post. A page sits in a browser. An email sits in an inbox.
- 2The rest of the campaign. Post, story, reel, page, email. Click one to bring it up. Pinch out to see them all at once.
- 3The ask box. Type, like ChatGPT. Or hold the space bar, point and talk.
- 4Library. Your products, photos, footage and logo. Drag anything in.
That's the whole screen. No canvas to arrange. No toolbars. No separate editors.
Today vs this
A canvas of cards and a long chat. Every edit adds a card. You hunt for the latest.
One piece, big. The campaign in a row under it. One ask box. Edits change the piece itself.
You describe what you mean in words.Remove the French here. Remove the French here.
Ananda Elixirs
You point while you talk.Olivia lights up each thing you mean as you say it.
An edit can change things you didn't ask for.this one change messed up too many things
PYM, Oct 23
Only what you named changes.Olivia ends with one line: what changed, and "nothing else changed."
Every edit is a new card. You hunt for the latest.which one is the most recent version?
2 Dogs Treats, twice in 20 minutes
Edits update the same piece.The latest is always the one on screen. Older ones sit underneath.
Photo, video, page and email each work their own way.based on the landing page design. give me an email design that leads people to…
PYM, Oct 22
Same screen, same moves for all four.Change the price once and every piece follows.
When a chat goes wrong, you start a new one and lose your decisions.I don't want to restart.
You, on the Kaphryn call, Nov 3
Decisions live on the piece, not the chat.Nothing to restart. "Redo this clean" keeps every decision.
Why we think this is right. We read 2,176 owner messages from the last 30 days. The median is 73 characters. Owners don't write briefs. They point and give short orders, the way they would to a designer sitting next to them. This UX is built for how they already talk.
First open to shipped, in 20 minutes
Maya runs Sunday Soda (made up) with one other person. She knows Shopify, Instagram, Canva and ChatGPT. She has never opened Figma. Here is her first morning.
Olivia does the setup. Maya says what she's making.
She pastes her store link, sundaysoda.com. In about a minute Olivia brings in her products, photos, logo and colors. The first screen is never empty. It shows three ready starts, each with a real first draft. Maya taps the first one, or just says "Blood Orange launch, next Friday." That's her first move.
The first edit teaches the one move.
The launch post opens big, as an Instagram post. She holds the space bar, puts her pointer on the can and says "make this a bit bigger." She moves to the label: "and never touch this label." Both light up while she talks. She lets go. The can grows. The label stays the same, pixel for pixel.
Can 12% bigger. Label untouched. "Always keep the label as it is?" Yes / Not now
Video uses the same move.
She says "make it move." Olivia drafted the shot plan while she worked, for free. Making the video costs credits, so it waits for her one tap. The reel plays on a phone, with sound. Its shots sit in a strip underneath, like the trim bar on an iPhone. She points at shot 4: "this one's slow, cut it." Gone.
Pages and emails show up where customers see them.
She clicks Page in the row. It opens in a browser, with the phone view beside it. She clicks both buttons and says "round, like before." Then Email. It shows up in an inbox, because the subject line is what customers see first. She points at it: "shorter, keep the 25%."
One change, every piece.
Her partner texts: it's 25% off now, not 20. She pinches out to see the whole campaign and says "it's 25 off now. everywhere." The number changes on all five pieces. Nothing else moves.
Olivia sets it up. Maya presses the button.
"Email Thursday at 9. Reel Friday." Olivia gets both ready. Maya clicks Send and Post herself. Olivia never sends or posts on its own.
Next week she starts a new campaign. Her rules and taste come with her. Nothing to explain twice.
The whole UX is six rules
The same six for photos, video, pages and email. No menus to learn.
-
Hold to talk. Point while you talk.
Hold the space bar or the mic button, talk, let go. Each "this", "here" or "these" means what your pointer was on when you said it. Rather click? Click a few things first, then talk or type.
Click both buttons, then type "round, like before."
-
See what Olivia understood, before it acts.
Each thing you mean glows with a number while you're still talking. If it's the wrong one, fix it before you let go.
The label lights up, but you meant the whole can: "no, the whole can."
-
What you see is what changes.
One piece open? Only that piece. Whole campaign on screen? All of it. Pointing at a part? Only that part. Everything else stays exactly as it was.
Every change ends with one line: "Changed: 2 buttons, now round. Nothing else changed."
-
Every edit updates the same piece.
No new cards. The latest is always on screen. Press and hold the piece to see the one before. Nothing is ever lost.
Want options? "Show me three." Pick one. The others stay in its history.
-
Say it once.
Say "always" or "never", or repeat yourself, and Olivia asks once: "Make that a house rule?" Facts like price and dates are stored once per campaign, so every piece agrees.
"Never change the text on the can." Now it holds for every photo.
-
Olivia is one step ahead.
Keep a photo, and the story size, email header and page hero are already drafted. Free things just happen. Anything that costs credits, like video, waits for your tap.
You never ask for a resize. It's there when you look.
How "this" works
Olivia hears each word with its exact time. It also knows what your pointer was on at that moment. So each "this" gets its own target, even two in one breath.
"make this a bit bigger…"
- Pointer was on
- the can
- Olivia does
- can 12% bigger
"…and never touch this label"
- Pointer was on
- the label
- Olivia does
- locks it, pixel for pixel
Questions you'd ask
How does Olivia know what I pointed at inside a photo?
The same way an iPhone lets you lift a dog out of a photo. Olivia cuts out the object under your pointer, in about a second. If it picks the label and you meant the can, say "the whole can".
Pages and emails are easier. They're built from real parts, so Olivia knows every button and heading exactly. In a video, it's the shot that was playing, at the spot you pointed.
What happens to the canvas we have?
It stops being a place you arrange. Its one real job, seeing a whole campaign at once, moves to the campaign view: pinch out and Olivia lays every piece out in a row.
We get there in steps. First, the canvas stays and any card can open big in its real frame (build 3 below). Once most edits happen there, the canvas becomes the campaign view.
And the chat?
The ask box stays, at the bottom, like ChatGPT. Olivia's answers show up as one line by the piece, and its questions sit on the piece too. Swipe up for the full chat if you want it.
But the chat stops being the memory. The piece is. Its versions, rules and references live on it. So there's never a reason to start a new chat. When something keeps going wrong, Olivia offers: "Redo this clean from your originals, keeping your 6 decisions?"
How do versions work?
Each piece has one strip of versions underneath it. A new edit goes on top and becomes what you see. Press and hold to compare with the one before. Click any old one, or say "go back to this morning's", to bring it back. Nothing is deleted.
Options work the same way. "Show me three" puts three side by side. You click one. The other two stay in the strip.
What about reference photos?
Drag one in from the Library, or paste a link. Point at what you want from it and say it: "this light", "this exact can", "just the idea". It pins beside the piece with that word on it. Click the word to change it.
Letters A, B and C stay, because owners already file things that way.
How does it feel on a phone?
The same screen, smaller. Your finger is the pointer.
Tap what you mean and it glows. Then hold the mic with your thumb and talk. Swipe left and right to move between the post, reel, page and email. Pinch out to see the whole campaign. Press and hold a piece to see the one before.
The phone may be the better fit. Pointing with a finger is more natural than with a mouse.
Do I have to talk? What about the camera?
No. Typing does everything talking does. Click a few things, then type "make these round". Talking is just faster when you're pointing.
The camera stays off. Pointing already says what you mean.
Where do house rules and taste live?
On two short pages, in plain words. House rules are your brand's dos and don'ts. Taste is what you tend to keep, learned from what you pick and the words you use. Cross out any line to fix it.
Rules are defaults, never walls. A direct ask still wins.
Can we build it?
Mostly, yes. Most pieces already exist, spread across four editors. The work is pulling them into one screen with one set of moves.
Ready now
The tech exists. It's product work.
- Words timed to the pointer. Takes, our talk-and-draw recorder, already logs each word, hover, click and stroke. It just waits until you stop, instead of lighting things up live.
- Exact highlights on pages and emails. They're HTML, so we know every part.
- Versions on the same piece. Version stacks and the Versions panel exist.
- "Only that section" on pages and emails. Section lock is built.
- "Make that a house rule?" Brand memory already asks Save / Not now.
- Real frames: post, phone, browser, inbox.
Works, but rough
Doable now. Needs care at the edges.
- Finding the can or the label inside a photo. Cut-out models do it in about a second. Small text and busy scenes are rough.
- "Only that part" in photos. Edit inside the area, paste the original pixels back outside it. Drawn Takes masks already route this way.
- One fact, every piece. Every piece has to read price and offer from one place.
- Changing one shot in a video.
Not yet
Not reliable today.
- Olivia talking back like a person, at a sane cost.
- Photos built as real layers, so any part can move.
- Changing one object inside a moving shot.
- Guessing which version you'll keep. Our creative model is right about 62% of the time.
Two honest notes.
The talk key is the space bar (when you're not typing), plus the mic button. Browsers don't reliably see the Fn key. A Mac app could use Fn later.
"Only that changes" holds because of how the edit is made, not because something checks afterward. Pixels outside the area are pasted back. Locked sections never reach the model. That fits your Oct 14 "no checks after generation" call. It's also the useful part of the Bend idea: Bend 2 proves rules about code, but it can't look at a photo, so we borrow the idea, not the language.
The first three builds
Each one ships on its own, on today's canvas, and each makes the next easier.
Point and talk, live
Make Takes the normal way to talk: hold space or the mic, no mode button. Light up each "this" with a number while the owner is still talking. Pages and emails first, because we know their parts. Photos next.
Also fixes multi-select (#1070) and the mic start (#1079).
Done when Kaphryn's "these buttons, and this underline" lands right the first time, on desktop and phone.
Edits change the piece, and only that
Every edit becomes a new version of the same card, latest on top, hold to see before. Turn on section lock for named edits. Named photo edits change only that area. Every change ends with the one "what changed" line.
Also fixes the stale page (#1074) and drifting edits (#1076).
Done when Ananda's "fix the logo, don't touch the text" keeps the text pixel for pixel, and nobody asks which one is latest.
Open it big, in its real frame
Double-click any card to open it big: a post as a post, a reel on a phone, a page in a browser beside the phone, an email in an inbox. Same moves on all four. The canvas stays behind it for now.
Starts the move off the canvas without breaking anything people use today.
Done when an owner makes a photo, a reel, a page and an email in one sitting, without a separate editor.
After these: house rules and references with a word on them. Then one fact across every piece, which is when the canvas becomes the campaign view.
Your calls
- Is this the direction: the piece opens big and that's where people work, and the canvas becomes the campaign view? I'd say yes.
- Should the first three go in the ledger as features? I'd file 1 to studio, 2 to outputs and studio, and 3 to outputs and video.