Prompting
Rothko
A second cursor on the board you can ask what to use. It opens the card filled in, and you decide whether to run it.
There are more tools here than anyone holds in their head, and the one that fits is often the one you have not opened. Rothko is the one you ask. Switch it on with the **@rothko** switch on the bottom bar and a second cursor appears beside your own, a little behind it, following you around the board, and a box opens at your pointer. Type, press Enter. The box steps aside while Rothko works and comes back when it is done, and `/` puts your caret back in it. Ask it in the words you would use with a colleague who knows the toolset. "How do I get this product shot onto a white background", "what would restyle this without losing the type", "I have a sketch and I want a model out of it." You do not have to name a tool, and naming one is fine when you already know. Select the picture you mean first and it reads that one. ## What comes back A bubble at the cursor says what it is doing while it thinks. When the answer is words, they stream into it so you can start reading before it has finished, and once it is done the answer is left on the board as a sticky note you can move, edit or delete. When the answer is "use this tool", the cursor leaves your side, opens that tool's card on the canvas and fills it in, then comes back to your pointer. Beside the card it leaves a sticky note saying what it did: the card it opened or changed, what it put in and set, and anything you still need to do before Run. It speaks when you ask, with one exception. On a card you opened yourself, once you have dropped something in and written the prompt, the cursor may switch the card's model to the one it thinks fits and say so in one line. It does that once per card, so a model you switch back to stays yours. A question it needs answered, a failure or a stop stays in the bubble until you ask again, click the board, or press Escape. Escape mid-answer stops it. ## The card opens, and then it stops The prompt is already written into the card, the images you were talking about are already in the slots, and the model it named is already on the pill. You can edit every one of those before anything happens. **It proposes. The Run button is still yours.** Nothing Rothko does spends credits, because nothing Rothko does runs. The price appears where it always appears, on the Run button of the card in front of you, and it is still the first moment anything is charged. If the card it wants is already open on your canvas, it changes that one instead of opening a second beside it. ## The tools it is allowed to name Rothko can only offer tools that are actually on your rail. It reads the live list every time you ask, so a tool that has been hidden or retired is not something it can suggest, and a tool that shipped this morning is available to it immediately. It names the model as well as the tool. A recommendation reads as a specific pill on a specific card rather than a category, which matters because most tools carry several models and they are not interchangeable. :::note[It reads the board, not your intention] Rothko proposes against what is actually on the canvas, which is why its suggestions are specific rather than generic. It is also the limit: a reference you meant to drop and did not is a reference it cannot see. ::: ## Watching it work, or not By default the cursor travels: you see it open the card, set the pill, write the prompt. If you would rather the card simply appear prepared, turn off **Rothko shows its work with the cursor** in Settings, under **Plnty AI** (<kbd>Ctrl</kbd> <kbd>,</kbd> opens Settings). Reduced-motion systems get that behaviour without asking. ## Teaching it Teaching starts from a result that missed. When you run a card Rothko prepared, a thumbs up and a thumbs down appear beside what it makes, on accounts where rating is switched on. A thumbs down records your vote and, on a board you can edit, opens a lesson with a picture of that result already in it as the first example and the question you asked attached. From there **Mark another** lets you drag a box over more of the board, such as the setup you started from, up to five pictures with a note on each. Rothko reads them with the runs you made on the board and drafts a lesson: what happened, when it applies, which controls did it. That draft is not the lesson. Your own sentence is the part worth writing, and then you confirm it or throw it away. A confirmed lesson teaches Rothko for you at once and goes into a queue for everyone else: somebody checks it against the run that actually happened before it reaches the knowledge every future answer reads from. A lesson that says a model did something the logs say it never did gets corrected or dropped. That check is the whole reason teaching is worth doing. One wrong fact folded in would be repeated back to you confidently for months. ## Where it is still rough It thinks before it speaks, so there is a pause before the first words arrive. The answer then streams quickly, but the wait at the front is real and we have not made it shorter. It can name the wrong tool. It is reading a board and a catalog, not your brief, and a plausible recommendation is not always the right one. Read the reason it gives you before you press Run, because the reason is usually where a bad suggestion shows itself. It has no taste of its own. It knows what the tools do, what this studio has actually reached for, and what is on your canvas. Deciding what the work should become is still the job you came here to do.There are more tools here than anyone holds in their head, and the one that fits is
often the one you have not opened. Rothko is the one you ask. Switch it on with the
@rothko switch on the bottom bar and a second cursor appears beside your own, a
little behind it, following you around the board, and a box opens at your pointer.
Type, press Enter. The box steps aside while Rothko works and comes back when it is
done, and / puts your caret back in it.
Ask it in the words you would use with a colleague who knows the toolset. “How do I get this product shot onto a white background”, “what would restyle this without losing the type”, “I have a sketch and I want a model out of it.” You do not have to name a tool, and naming one is fine when you already know. Select the picture you mean first and it reads that one.
What comes back
Section titled “What comes back”A bubble at the cursor says what it is doing while it thinks. When the answer is words, they stream into it so you can start reading before it has finished, and once it is done the answer is left on the board as a sticky note you can move, edit or delete. When the answer is “use this tool”, the cursor leaves your side, opens that tool’s card on the canvas and fills it in, then comes back to your pointer. Beside the card it leaves a sticky note saying what it did: the card it opened or changed, what it put in and set, and anything you still need to do before Run.
It speaks when you ask, with one exception. On a card you opened yourself, once you have dropped something in and written the prompt, the cursor may switch the card’s model to the one it thinks fits and say so in one line. It does that once per card, so a model you switch back to stays yours. A question it needs answered, a failure or a stop stays in the bubble until you ask again, click the board, or press Escape. Escape mid-answer stops it.
The card opens, and then it stops
Section titled “The card opens, and then it stops”The prompt is already written into the card, the images you were talking about are already in the slots, and the model it named is already on the pill. You can edit every one of those before anything happens.
It proposes. The Run button is still yours.
Nothing Rothko does spends credits, because nothing Rothko does runs. The price appears where it always appears, on the Run button of the card in front of you, and it is still the first moment anything is charged.
If the card it wants is already open on your canvas, it changes that one instead of opening a second beside it.
The tools it is allowed to name
Section titled “The tools it is allowed to name”Rothko can only offer tools that are actually on your rail. It reads the live list every time you ask, so a tool that has been hidden or retired is not something it can suggest, and a tool that shipped this morning is available to it immediately.
It names the model as well as the tool. A recommendation reads as a specific pill on a specific card rather than a category, which matters because most tools carry several models and they are not interchangeable.
Watching it work, or not
Section titled “Watching it work, or not”By default the cursor travels: you see it open the card, set the pill, write the prompt. If you would rather the card simply appear prepared, turn off Rothko shows its work with the cursor in Settings, under Plnty AI (Ctrl , opens Settings). Reduced-motion systems get that behaviour without asking.
Teaching it
Section titled “Teaching it”Teaching starts from a result that missed. When you run a card Rothko prepared, a thumbs up and a thumbs down appear beside what it makes, on accounts where rating is switched on. A thumbs down records your vote and, on a board you can edit, opens a lesson with a picture of that result already in it as the first example and the question you asked attached. From there Mark another lets you drag a box over more of the board, such as the setup you started from, up to five pictures with a note on each. Rothko reads them with the runs you made on the board and drafts a lesson: what happened, when it applies, which controls did it.
That draft is not the lesson. Your own sentence is the part worth writing, and then you confirm it or throw it away. A confirmed lesson teaches Rothko for you at once and goes into a queue for everyone else: somebody checks it against the run that actually happened before it reaches the knowledge every future answer reads from. A lesson that says a model did something the logs say it never did gets corrected or dropped.
That check is the whole reason teaching is worth doing. One wrong fact folded in would be repeated back to you confidently for months.
Where it is still rough
Section titled “Where it is still rough”It thinks before it speaks, so there is a pause before the first words arrive. The answer then streams quickly, but the wait at the front is real and we have not made it shorter.
It can name the wrong tool. It is reading a board and a catalog, not your brief, and a plausible recommendation is not always the right one. Read the reason it gives you before you press Run, because the reason is usually where a bad suggestion shows itself.
It has no taste of its own. It knows what the tools do, what this studio has actually reached for, and what is on your canvas. Deciding what the work should become is still the job you came here to do.