Talking character
An animated cast (a person, animal, creature, object or ingredient given a face) talks to camera: each one says who it is, what it does and what that changes for the viewer, and every line is staged literally.
Use it in three steps
- 1
Connect roasOS
Add the MCP server to your agent and sign in. MCP is included on every plan.
claude mcp add --transport http roasos https://mcp.roasos.com/mcp - 2
Add the skill
Save the file in the skills folder of your agent. This command is for Claude Code.
mkdir -p ~/.claude/skills/roasos-talking-character && curl -o ~/.claude/skills/roasos-talking-character/SKILL.md https://www.roasos.com/formats/talking-character/SKILL.md - 3
Ask for the ad
Your agent plans the ad and shows each price. Nothing spends until you say yes.
Make a talking character ad where an animated character sells my product.
The structure of the ad
Ingredient stories and products that are hard to show in use.
- 10–3 s hook: a first frame that stops a muted scroll, and a first line that names the viewer's problem, said by the character or the villain.
- 23–8 s the character introduces itself and its job, acting the line out as the world reacts.
- 38–12 s what that changes for the viewer, one effect you can see, from the product facts; a handoff when there is a second character.
- 412–15 s the pack, whole and label readable, makes the ask.
How to shoot it
One look line opens every prompt word for word; one reference sheet per character rides in every render; a non-living character has a face clear of the label and a way to gesture; every shot stacks at least two of camera move, character action and a world reaction; native speech with the emotion written in, separate speech and lip-sync only for hard words.
What breaks it
A still head on a still set, a face or limbs over the label, a pack reshaped into a toy, a voice that changes between shots, effects that do not illustrate the line, words crammed past about 2.5 per second.
Read the file
---
name: roasos-talking-character
description: Plan and make a talking character with roasOS over MCP. Use when the user asks for a talking character, a cartoon spokesperson or ingredient characters.
---
# Talking character
An animated cast (a person, animal, creature, object or ingredient given a face) talks to camera: each one says who it is, what it does and what that changes for the viewer, and every line is staged literally.
## 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
An animated performer looks down the lens and makes the case. Who it is, how many there are, the look and the arc are all your call. An animated person with a sharp line beats a clever ingredient with nothing to say.
### Cast
People, animals and creatures already perform: render them in the look. Anything else (pack, ingredient, organ, molecule, household object) has to be built into a performer: eyes, a mouth, some way to gesture (stubby arms, a bending neck, a lid that flaps). Choose where the face sits on that body, clear of any label: on the cap, on a blank panel, above the print. A real pack stays the real pack; the photo outranks the design.
In an ensemble each character gets one job and one voice. The pack usually comes last and makes the ask.
### One sheet per character
Make the sheets before any shot (front, side and back views on one image). Stack the design: look and subject with a few adjectives, body, colour and surface, face, limbs, one pose with intent. Go one notch past the obvious so it sticks. The brand's palette and light live in the character, which frees the setting to change line by line; only a villain or an ugly before state breaks from it.
Write one look line (medium, material, light) and open every prompt with it word for word. A shot prompt then gives only place and action; the sheet owns the appearance. Without a sheet the character mutates by shot three.
### Script
The unit is three moves: my name, my job, what changes for you. Three arcs:
- Roll call: no problem framing, each character presents itself, the pack closes. Premium brands, ingredient line-ups.
- Pain first: a narrator or the product names something the viewer lives with, then the character arrives. Cold audiences.
- Villain flip: the problem speaks first, proud of the damage; a hard turn; the hero answers in the same form. Most entertaining, easiest to overdo.
Hold one register throughout (a weary night-shift fixer, a smug little specialist). Write effects you could film, only what the product facts back. Two hooks: a first frame that stops a muted scroll and a first line that names the viewer's problem.
### Picture is the proof
Stage each sentence literally. When the character says it closes a gap, the gap closes under its hands. Effects that illustrate nothing are waste. Sound off, the ad should still make sense.
### Motion
Heads never sit still. Stack at least two of three in every shot: the camera travels (push, orbit, whip, tilt); the character acts (enters from a named side, leans, performs the line); the world answers (light shifts, particles fly, walls respond). A handoff is choreography: one shoves, tosses or follows the other into frame.
### Voice
Default to native speech: the line quoted in the take, the speaker named the same way each time, the emotion written in ("the oat clump, flat and unimpressed:"), one exact voice description per character in every take.
Coined names, long ingredient words and strings of numbers get mangled. For those shots, record the line with `roasos_generate_audio` (a voice from `roasos_list_voices`, hard words spelled by sound in the voice input only), then pass it as an audio reference to a take on the reference-media role, which lip-syncs to it.
### Length
Count words: about 2.5 per second, rounded up to the model's next allowed length, so 15 s holds about 35. The line sets each shot's length. No monologues.
### Order
1. Sheets on approval cards (quote and confirm); one yes.
2. Script, one line per shot, into a storyboard table.
3. Storyboard sheet: each panel that shot's place and action, every cast sheet and the product attached. Approve it.
4. Compile the board into one video prompt, then one video card (quote and confirm); name each reference by role, each character by name.
5. Check face, label, every word. `roasos_captions_auto_subtitle` last.
## Format data
- Aspect ratio: 9:16
- Default length: 15 seconds
- Needs: a product
- Example request: Make a talking character ad where an animated character sells my product.
Beats:
- 0–3 s hook: a first frame that stops a muted scroll, and a first line that names the viewer's problem, said by the character or the villain.
- 3–8 s the character introduces itself and its job, acting the line out as the world reacts.
- 8–12 s what that changes for the viewer, one effect you can see, from the product facts; a handoff when there is a second character.
- 12–15 s the pack, whole and label readable, makes the ask.
How to shoot it: One look line opens every prompt word for word; one reference sheet per character rides in every render; a non-living character has a face clear of the label and a way to gesture; every shot stacks at least two of camera move, character action and a world reaction; native speech with the emotion written in, separate speech and lip-sync only for hard words.
What breaks it: A still head on a still set, a face or limbs over the label, a pack reshaped into a toy, a voice that changes between shots, effects that do not illustrate the line, words crammed past about 2.5 per second.
Ship 10x more creatives. Spend 90% less.
Product photos, video ads, and copy. All from one platform, ready in minutes.
Get started