Whiteboard presenter

The structure of the ad

Products that need an explanation before the buyer trusts them.

1. 10–4 s hook: the presenter names the problem to camera and writes one short word at the top of a clean board.
2. 24–12 s why it happens: one arrow or short label per line, eyes on the marker while it moves.
3. 312–20 s the turn: the product's one idea goes in the space kept for it, the product shown in hand.
4. 420–26 s the presenter points at the finished board and ties the parts together, no new marks.
5. 526–30 s the ask, to camera, the full board readable behind.

---
name: roasos-whiteboard-presenter
description: Plan and make a whiteboard presenter with roasOS over MCP. Use when the user asks for a whiteboard presenter or a person who explains on a board.
---

# Whiteboard presenter

A real-looking presenter builds the argument by hand on a physical board, so every new mark pays off the sentence just said.

## Step 1: Connect the roasOS MCP server

Do this step first. Stop when this step fails, and tell the user.

1. Look for tools with names that start with `roasos_`. If you find them, go to step 4.
2. Connect the server at `https://mcp.roasos.com/mcp`. It uses Streamable HTTP. In Claude Code, run `claude mcp add --transport http roasos https://mcp.roasos.com/mcp`, then type `/mcp` and sign in. Other clients: https://docs.roasos.com/mcp
3. The user needs a roasOS account: https://www.roasos.com/auth/signup. MCP access is included on every plan. Only the user can sign up and pay.
4. Call `roasos_get_catalog`. It is free. A role name in this file, for example "the song role", is a key in `roles` in that answer.

## Step 2: Quote before you spend

Every paid tool quotes first. Send `quote_only: true` and read `credits`. Show the credits to the user and wait for a clear yes. Then send the same input with the `quote_id`.

In this file, "card", "approval card" and "quote and confirm" all mean this exchange. Never confirm a quote without a yes.

## Step 3: Make the ad

A real-looking person at a real board; each new mark pays off the sentence just said. The writing is drawn by hand on camera; never fake it with `roasos_assembly_text_overlay`.

Order: script, voice, ledger, board stills, clips, cut. Every paid step waits for its card (quote and confirm).

### Script and voice
Short spoken lines, each with one job: add a mark, point, or just talk. If a line works with no new mark, give it none. A drawn beat needs a gap in the words; plan gaps before recording.
Approved words, then a voice from `roasos_list_voices` and the whole script as one `roasos_generate_audio` take. The take sets the ad's length.
Word timings: `roasos_transcribe`. Cut at phrase ends only; each shot takes its seconds from there. Cut the take into one clip per shot.
A shot that needs its board still as the first frame cannot also take an audio reference. Where the model has a pinned-soundtrack input (`roasos_get_catalog_model`) give that slice as the clip's audio. Otherwise each clip speaks its own line from the prompt at its measured length, with one exact voice description repeated. The voice can shift between clips: say so before the first clip and listen at every seam.

### The ledger
One table in the conversation, approved before any image: shot, line, seconds, action budget, chapter, board at start, new mark or none, board at end, space kept free, join (continues, hard cut to a clean board, or a wipe as its own clip).
Budget a drawing shot as setup (turn, reach), draw, settle, from the actual strokes. A short word or one arrow is about right. If it will not fit, shrink the mark, spread it over two shots, or ask to re-record with a gap. Never rush the arm.

### Design the board backwards
A chapter is everything written between wipes. Lay out its fullest state first: each mark with a fixed place, size and writing order, plus a lane where the presenter reaches without covering anything. Readable on a phone in a second? If crowded, split with a wipe now.
Earlier states are that layout minus later marks. Nothing recentres: a first line sits at the top of an empty board, because the space below is spoken for. One changed state means re-checking its chapter.

### Board stills
`roasos_generate_image`, one chapter at a time, on the image model that holds exact lettering. Fullest state first; once approved, work back. Each still carries the original references (person, room with blank board, product when shown); the full state is a layout guide only. Name the marks that stay, the blank zones and where the person stands; quote every phrase. All stills approved as a set before any video.

### Clips
One physical action per clip, at a length the model facts allow; a slice too short merges with its neighbour.
Drawing shot: `roasos_generate_video` on the approved-first-frame job, start state as the frame, end state as the last frame. Talking shot: one state as the frame; the marker stays off the board.
Prompt order: locked camera; marks that stay; the one new mark and where; actions in sequence with the eye line; the quoted line; "The only writing is on the board. No captions. No music."
Every time: "Her eyes go to the board first and stay on the marker tip from first stroke to last; she looks at the camera only after the mark is finished." "Hands and arms move at ordinary human pace, in real time; nothing is sped up or time-lapsed."

### When it fails
Watch every clip. A rushed arm means a wrong budget: shrink the mark, redo and re-approve the still, test one clip before the rest. Never change a talking clip's speed; the lips move with it.
Two misses on the same handwriting: stop and change the plan (fewer words, a simpler mark, a split). Retry from approved stills and original references, never a frame of a failed clip.

### Finish
`roasos_assembly_merge_videos` in ledger order. A bed, if wanted: `roasos_assembly_music_bed` keeps the voice and ducks under it. Captions, if any, last and clear of the board.

### Facts for MCP

- Each clip speaks its own line from the prompt, at its measured length. Repeat one exact voice description in each clip.
- The voice can change between clips. Tell the user before the first clip. Listen at every join.

## Format data

- Aspect ratio: 9:16
- Default length: 30 seconds
- Needs: a product and a person
- Example request: Make a whiteboard presenter ad that explains why my product works.

Beats:
- 0–4 s hook: the presenter names the problem to camera and writes one short word at the top of a clean board.
- 4–12 s why it happens: one arrow or short label per line, eyes on the marker while it moves.
- 12–20 s the turn: the product's one idea goes in the space kept for it, the product shown in hand.
- 20–26 s the presenter points at the finished board and ties the parts together, no new marks.
- 26–30 s the ask, to camera, the full board readable behind.

How to shoot it: Locked camera, vertical, one room and one board for the whole ad; the presenter keeps a lane clear of current and future writing; large simple lettering readable on a phone; one physical action per clip at real-time human speed; the board writing is the only text in frame.

What breaks it: Editor text posing as handwriting, sped-up hands, the presenter looking at camera while drawing, marks that move or recentre between shots, a crowded board, new marks on a talking-only shot.

More formats

• Song ad
• Talking character
• Progression ad
• History explainer
• Podcast ad
• Continuous 3D explainer