Skip to main content
Open the deck whose master you want the output to inherit (your firm’s template, or any deck already on brand) and describe the whole thing in one message. Pretty writes a plan, then works down it one slide at a time.

​
What happens after you send a brief

1

It plans

A Planning steps row appears with a task list, roughly one task per slide, named by what that slide has to do. Tasks move through pending, in progress, completed.
2

It builds one slide at a time

For each task you will see the same short cycle: a PowerPoint row writing shapes into your deck, a Snapshot row where Pretty looks at what it just made, and often a Reviewing design row where it critiques the result.
3

It fixes before moving on

Overflowing text, a collision, a column that does not line up. Pretty catches most of these on its own snapshot and corrects them before starting the next slide.
Because it finishes each slide before starting the next, you can read the deck as it lands rather than waiting for the whole run.

​
What to put in the brief

The plan is only as good as the brief. Five things carry most of the weight.
If you already have approved slides for this kind of deck, upload them as a slide pool first. Pretty will search those layouts and reuse them rather than inventing a structure from scratch.

​
A worked example

Copy two habits from that brief: name the repeated layout (“same layout each time”), and tell Pretty what to do when the source is ambiguous.

​
Steer it while it runs

You do not have to wait for the run to finish.
  • Send a correction now. Type and send while it is working. The message shows as Queued, and Pretty picks it up at the next step. Send now pushes it in sooner; Dismiss drops it.
  • Point at something. Select shapes on a slide it already built, then send. The composer shows a selection badge and the agent works on exactly those shapes.
  • Answer its questions. If the brief is ambiguous, an Asking you row appears with two to four options plus a free-text Other. Answering is faster than letting it guess.
Useful mid-run corrections:

​
Stop and restart

Stop generation in the composer halts the run. Slides already written stay in your deck. Pretty only rolls back a step that failed part-way, not work that finished. To restart, send a new message in the same chat. Pretty still has the plan and the slides it built, so “carry on from task 7” works. Start a New Chat instead when you want a clean slate, or when the deck has changed direction enough that the old plan is noise.
Long runs use a lot of context. Pretty compacts the conversation automatically. You will see Compacting… and a note that older messages were summarized. See Context and compaction.

​
Where the content comes from

A brief with no source material gets you structure and prose, not facts. Two ways to ground it:

Your own documents

Attach the board pack, the transcript, the model. Pretty reads them and builds from what is actually in them.

Live web research

Ask it to research a market or a competitor and it adds a Sources: footer to the slides that came out of that research.