Editor assistants
The AI Help panels inside each editor: what each one builds, and how to keep or undo what it did.
Most editors have their own assistant. Describe what you want, and it edits the open character, scenario, collection or memory book. Open it with the sparkle at the end of the editor's tab bar (AI Help), or the sparkle in the page header on a phone. The quick search's Open This Editor's Assistant does the same.
These assistants work on the page in front of you. For questions about Voxta itself, from anywhere in the app, use the App Assistant.
The assistants run on the Assistant service of your configuration (Settings › AI Setup › Configuration), or on Chat Replies when none is set. They work best with a model that can call tools and see pictures. Reference material they need is loaded as skills only when a task calls for it.
Checking their work
- Changes go straight into the editor. You see them as they happen, unsaved edits included.
- Each reply is validated. When a reply leaves the resource with errors it didn't have before, the assistant is sent those errors and fixes them in a follow-up turn.
- Revert on a reply rolls the resource back to before that reply and removes it. Revert the newest one first.
- Read what it wrote before you publish. An assistant can be confidently wrong about a character's story or a script's logic.
Scenario editor
| Assistant | What it does |
|---|---|
| Scenario (Scenario AI Help) | Answers questions about the scenario (events, buttons, actions, flags, scripts) and builds or changes them. Sees the editor state and your assets. Create with AI in the new-scenario dialog opens a blank scenario with it ready. |
| Button Stylist | Restyles the scenario's buttons to a look you describe (retro RPG, sci-fi HUD…), creates new ones, or generates icon artwork for them. |
Character editor
| Assistant | What it does |
|---|---|
| Character Builder | Writes the whole character from a description: looks, personality, backstory, first message, dialogue, memories and scripts. Ask for a change and it rewrites only that. |
| Emotes | Shapes how the character shows feelings: which emotes it has, when the AI picks each, how they look. |
| Avatar Folders | Sets up the poses and situations the character moves between: when the AI switches to each folder and what it knows while there. |
| Reply Suggestions | Sets up the tones the player's ready-made replies are written in, each in its own voice and color. In a chat, suggestions are only offered when the text service supports structured output. |
| Expressions | Looks over one avatar folder at a time and fixes pictures that are inconsistent with the rest. You approve or reject each change. |
The expression generator page has its own AI Help with the Avatar Folders and Expressions assistants, and a Prompt assistant setting that writes each expression's image and video prompts.
Collection editor
The Collection assistant (Collection AI Help) works on a collection. It can:
- name and describe it, and write its readme,
- arrange its shelves,
- add or remove characters and scenarios by searching your library,
- set its gallery folder,
- lock a card behind a "Coming Soon" or a link.
It reads the members' cards before writing about them, but can't change a character's own card. When one needs fixing, it offers a button that opens that character in the Character Builder with the request already written.
Memory book editor
The Skills Assistant writes and maintains a book's skills and memories from documentation, answers questions about the book, and tests what a question would recall. See Memory books.
Prompt assistants
The Create pages have assistants that write prompts for the model you're using.
Image (Open the Prompt Assistant):
| Assistant | For |
|---|---|
| Tag Prompts | Tag models like Pony, Illustrious and SDXL: tags and a matching negative prompt. |
| Natural Prompts | Models that read sentences, like Flux and Qwen Image: a rich, detailed prompt. |
| MOSS-Image | MOSS-Image: a detailed description with exact text to show, or a laid-out poster, sign or panels. Attach a picture and it writes the prompt that recreates it. |
| Image Edit | Editing models like Qwen Image Edit: the edit instruction for the input images, or a new image of a character from a reference sheet. |
Any of them can start from a picture you attach, and writes straight into the playground's prompt.
Video (Improve my prompt, with Improve with picking the assistant):
| Assistant | For |
|---|---|
| H3 Image to Video | One MiniMax H3 prompt from a first frame and the motion you describe. |
| H3 Concept to Video | One MiniMax H3 prompt from several reference images (characters, outfit, setting), each kept distinct. |
| LTX Video | One LTX caption: picture, camera and soundtrack in a single paragraph. |
| Prompt Assistant | A clean, vivid prompt from a description or a prompt to sharpen, for any other model. |
A video module picks its own by default (H3 for MiniMax H3, LTX for LTX).
Music: the songwriter writes a song's title, style and lyrics, or rewrites the lyrics you have.
3D Model: the assistant button turns a few words into a full 3D description for the mesh model.