> ## Documentation Index
> Fetch the complete documentation index at: https://help.getatelier.design/llms.txt
> Use this file to discover all available pages before exploring further.

# Build a board with AI

> Assign a task to your AI teammate and it drafts a whole board, or ask AI Chat to compose the page in front of you. Everything stays editable afterward.

Not the AI Chat panel — this runs as a task. Describe the board you want,
assign it to your AI teammate, and it works in the background: it creates
the pinboard, sources images, arranges the page and writes labels. You
review and adjust from there, the same as any board you built by hand.

## How it works

<Steps>
  <Step title="Describe the board in a task">
    Create a task and put the brief in the description: the room, the
    mood, the pieces to include. A specific brief gets a better first
    draft — *warm neutrals, boucle and brass, a rug and a floor lamp* beats
    *living room ideas*.
  </Step>

  <Step title="Assign it to your AI teammate">
    Open the task's **Assignees** field — the AI teammate sits alongside
    the rest of the studio in the picker.
  </Step>

  <Step title="Atelier builds it">
    It reads the task, checks a library of well-composed layouts for
    inspiration, then creates the pinboard: images placed and sized,
    labels written, split across pages if there's a lot to show. Progress
    posts to the task as it goes, with a link to the pinboard when it's
    done.
  </Step>

  <Step title="Review and refine">
    Every item it places is a normal pin — drag, resize, relabel, delete,
    add more. Treat the draft as a starting point, not a finished
    presentation.
  </Step>
</Steps>

## Ask the chat for a room scene

With a pinboard open, **AI Chat** builds pages too — and it can lay pieces
out as a room rather than a grid: the rug on the floor, the chair standing
on it, the chandelier hanging above, everything at its real relative size.
Ask for that and you get a styled vignette; ask for a product showcase and
you get an even grid.

Six things worth saying in the request. None are required — it will infer
what you leave out — but each one it doesn't have to guess is one less
thing to redo:

<AccordionGroup>
  <Accordion title="That you want a scene, not a product grid">
    Say *scene*, *vignette*, *as if it were a real room*, or place one
    piece against another (*the lamp on the side table*, *art above the
    bed*). Left unsaid, a set of pieces that furnish one room is read as a
    scene anyway — but a row of chair options stays a grid, which is
    usually what you want.
  </Accordion>

  <Accordion title="Which room or zone it is">
    *Living room*, *bedside*, *dining group*. This decides the stacking
    order — floor covering, then furniture, then what stands on the
    furniture, then anything on the wall or ceiling.
  </Accordion>

  <Accordion title="Which pieces to include">
    Name them, or just say *the products on page 3*. It picks the anchor
    itself: the largest floor piece, which everything else is then sized
    against.
  </Accordion>

  <Accordion title="Any scale or emphasis you care about">
    *Make the armchair the hero*, or *I want the rug to read as the base*.
    By default it sizes everything from the real dimensions in your product
    library, at one shared scale — so a 19-inch side table comes out about a
    fifth as wide as a 96-inch sofa, not the same size.
  </Accordion>

  <Accordion title="Whether you want labels">
    It does not label anything unless you ask. Say *label each piece with
    its name and price* if you want captions.
  </Accordion>

  <Accordion title="Which page">
    *Page 3*, or *a new page after the current ones*. Otherwise it works on
    page 1.
  </Accordion>
</AccordionGroup>

A request that covers all six: *On page 3, compose a living room scene with
the walnut armchair, the olive grid rug and the cloud chandelier — as if
they were standing in a real room, rug underneath, chandelier above. No
labels.*

<Tip>
  Scene pages come out best from **solo product shots** — a piece
  photographed alone. When a photo is a styled room shot, there is no
  background to remove, so the rest of that room comes along with it. The
  assistant will spot this and crop the surroundings away, but a clean
  cut-out image gives a better result with less work.
</Tip>

## Notes

* The **task** route above and **AI Chat** are different ways in: assign a
  task and it builds the whole board in the background; open a board and
  chat, and it edits the page in front of you, one request at a time.
* The board it produces is a first draft. It sources images and composes a
  layout; it does not know your client's taste, so check what it picked
  before sharing the board.
* Building a board spends AI credits, the same as any other AI task.