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

> ## Agent Instructions
> To build and serve your own model, start at /deploy/development/quickstart and /deploy/development/overview. Deploying is the default path: reactor init scaffolds a workspace, reactor auth login authenticates, and reactor model deploy registers the model, publishes the release with the weights/ folder, and activates it on Reactor's GPUs, in one command from that workspace. Docker must be running, because the publish step builds the image locally. Bump model.version in reactor.yaml before redeploying a change, because a release that already has an image is reactivated as it is. Deployment access is granted per account, so contact team@reactor.inc if a deploy is refused. Every key in reactor.yaml is documented at /deploy/platform/reactor-yaml. Model code imports reactor_runtime; Python client code imports reactor_sdk. The runtime overview explains the model interface. Running the model on your own machine with reactor run is optional and needs a GPU you attach with --gpus; /deploy/development/local-testing covers that loop and pairs a complete brightness model with a Python client test in a separate brightness-test workspace.
> Reactor hosts multiple models, each with its own connect slug (modelName) and command/event schema. The video model catalog — slug, typed SDK package, and links to its schema — is at /model-api-reference/overview. Robotics policy documentation starts at /robotics/overview; The launch policies are Cosmos Nano DROID (/robotics/cosmos/nano-policy-droid/), DreamZero DROID (/robotics/dreamzero/droid/), and FLUX Action DROID (/robotics/flux-action/droid/), organized by model family and embodiment. Some models expose one slug per experience (e.g. HappyOyster); always take the slug from the model's own pages, never guess it.
> Fastest path to a working app: `npx create-reactor-app my-app --model=<slug>` scaffolds a complete app with secure auth wired up. Typed TypeScript SDKs are published as @reactor-models/<model>; Python uses the base reactor-sdk package.
> Auth: exchange an API key (rk_...) for a JWT via POST https://api.reactor.inc/tokens from your server. Never put the API key in client-side code.
> Append .md to any docs URL for clean Markdown. Search these docs via the MCP server at https://docs.reactor.inc/mcp.

# PixVerse R2 Persona prompt guide

> Choose the image, write the persona and speaking style, pick a voice, and talk so the character hears you.

PixVerse R2 Persona gets its prompt in `create_character`, once for each character. The prompt is a
`name`, a `persona`, an optional `speaking_style`, and an optional `refusal_reply`. After that, you
can talk to the character with `say` or over the `mic` track. This page covers each of these inputs
and what to expect in a turn.

## Choose the image

The character looks like the image that you upload, so the image is the most important input:

* Use one clear, front-facing picture of one person or character.
* The model crops the centre of the image to `aspect_ratio`. The character looks like the cropped
  image, so keep the face and shoulders near the middle.
* Match the shape of the image to the call. Use a tall image for `9:16` (720 × 1280, the default)
  and a wide image for `16:9` (1280 × 720).
* Use PNG, JPEG, or WEBP, at most 20 MiB and 40 megapixels.

## Write the persona

The persona tells who the character is and how it behaves. It can have up to 4,000 characters.
Write what the character knows and what it does, as facts about it. Do not write stage directions,
because the character can read them aloud.

You are Ava. (Smiles warmly and tilts her head.) You are a photographer. \*pauses thoughtfully\*

You are Ava, a street photographer in an old town. You have shot its markets and alleys for ten
years. You love morning light, film cameras, and the people who let you take their picture. You
answer questions about photography in plain words and ask about the other person's photos.

Leave gestures to the character. The character makes gestures by itself and reports each one in a
`character_action` message.

The character does not speak first. If you want an opening line, send it with `say` after
`speech_ready` is true.

## Name the character

`name` has up to 20 characters. The character uses this name when it introduces itself, so use the
name that you want to hear.

## Set the speaking style

`speaking_style` sets the language, the tone, and the length of replies. It can have up to 1,000
characters. If you leave it blank, the character uses short, natural sentences in the language
that the user speaks. If you want a different style, set `speaking_style`:

Answer in English even if asked in another language. Warm and a little dry. One or two sentences
per answer.

## Pick a voice

`list_voices` returns every voice, with a `description` and the `languages` that it speaks. It also
returns `default_voice`. The default voice, `marin`, is a natural voice that speaks most languages.
The voices whose names start with `saturn_zh_` are Chinese character voices. They also speak
English, but with a strong Chinese accent.

Pass a `voice` from the list to `create_character`. To use the default voice, leave `voice` blank.
If the list does not have the voice, the model refuses `create_character` with `invalid_voice`.

## Set a refusal reply

The generation service applies its own content policy to the image, the persona, and what is said to
the character. When the service declines a turn, the character sends `input_declined`. Then the
character says its `refusal_reply` instead of an answer. If you leave `refusal_reply` blank, the
character politely changes the subject. You can also write up to 500 characters for the character
to say in its own voice:

Let's keep this about pictures. Want to hear about the best light in the old town?

## Talk to the character

You can reach the character in two ways. Both need `speech_ready` to be true:

* **`say`** sends one turn of text, up to 800 characters and 800 bytes of UTF-8 (about 260 Chinese
  characters). The reply is `said` with a `prompt_id`.
* **`mic`** carries your speech. Publish it before `start_call`. The character then answers your
  speech without a command.

Speak after the character finishes its answer. If you speak while it answers, it may not hear you.
A clear, natural voice works best.

## Common fixes

| Problem | Fix |
| - | - |
| The character reads out a gesture | Remove stage directions from the persona. The character makes gestures itself and reports them in `character_action`. |
| The character says nothing after the call starts | It waits for your first turn. Send a line with `say`, or speak into the microphone. To have it greet first, wait until `speech_ready` and `video_receiving` are both true. Then `say` a short line such as "Hello!". |
| Speech is not answered | Check `speech_ready` and `mic_forwarding` in `session_state`. The character does not answer speech before `speech_ready` is true. It does not hear a microphone that sends silence. |
| `say` is refused with `not_live` | There is no call to hear the text. Wait for `speech_ready`, or start a call. |
| `say` is refused with `text_too_long` | Keep the text under 800 bytes of UTF-8. Split a long message into turns. |
| The character answers in the wrong language | Name the language in `speaking_style`. |
| The character's face is cut off | Keep the face and shoulders near the middle of the image, in the shape `aspect_ratio` sets. |
| `input_declined` keeps arriving | The content policy declined the turn. Change the wording, or set a `refusal_reply` that fits the character. |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.