Avatars and Create with your AI assistant
Use your Mango9 MCP connection to inspect Create projects, prepare media work and reuse versioned avatars with the right inputs and permissions.
Connect once, choose the right project
Connect your assistant through Central, then ask it to list your available Create projects. Choose the project before requesting media work. Account-library actions such as saving or deleting an avatar require the appropriate owner/write access; a read-only connection cannot make those changes.
An avatar is a reusable identity record with approved references and optional voice material. Saving it does not generate a video, clone a provider voice immediately or start paid work. A later generation has its own input checks, quote and approval.
Prepare the avatar in the app
- Open Avatars & Characters and start a new avatar. Give it a recognizable name, choose its type and describe an invented character’s appearance.
- For a real person, enter the required identity details and follow the consent recording card. Read the exact current statement shown by the form.
- Upload the photo references and choose the clearest front-facing image. Wait for finished upload statuses and thumbnails. A video source still needs a front photo for its thumbnail and starting frame.
- Add optional videos where the form supports them, and choose the primary clip. Model limits govern accepted formats, count and duration.
- Choose the voice. For cloning, supply a clean voice sample and the required permission recording. The voice sample and consent have different jobs.
- Review the summary and choose Save. Confirm the avatar appears in the library with the expected photo. Uploaded files alone do not create the saved avatar.
Consent, voice and reference checks
Current voice routes use Mango9’s service-neutral permission wording for a new recording. An existing valid receipt can preserve earlier provider wording. Do not assume that every model needs a Google-specific recording, or that one statement is valid for every future route.
Changes to the legal name, avatar name, language, model or required statement can invalidate earlier consent. Re-record when the form asks. A recording stored successfully is not, by itself, a provider verification.
The current identity API accepts owned JPEG/PNG photos up to 10 MiB each. The app may convert or resize supported originals first. Voice samples are 3–60 seconds and up to 15 MiB in the supported audio formats. Read the selected form/model limits before uploading; requirements can differ between saving an avatar and generating media.
Ask the assistant to inspect before saving
The current Create MCP routes include listing and reading avatars, requesting a save, requesting a new version and requesting deletion. Tool discovery may group or prefix these names through the unified Mango9 connection; ask by goal rather than guessing an endpoint.
Show my saved avatars, their latest version and thumbnail.
For “Alex”, check the approved photo and any consent or voice requirements.
Explain what is missing before preparing a save or generation request.| Goal | Create tool name |
|---|---|
| List saved avatars | create_list_avatars |
| Read one avatar | create_get_avatar |
| Prepare a save | create_request_save_avatar |
| Prepare a new version | create_request_new_avatar_version |
| Read an exact version | create_get_avatar_version |
| Prepare deletion | create_request_delete_avatar |
MCP access does not replace the person’s consent or the model’s media requirements. The assistant must use available owned assets and valid permission records.
Use a saved avatar in generation
Motion control uses suitable reference motion as input; saving a primary avatar video does not automatically pass that video to every model. Use the model guide and the live model form to confirm exactly which references are required.
- Ask the assistant to inspect the available model and mode requirements for your goal. A talking portrait, image-to-video shot and motion-control clip accept different inputs.
- Choose the exact saved avatar version and required source assets. For Kling 3 named image elements, the current identity contract expects two to four approved images per element, up to three elements, with a starting image in frames mode. Do not assume a single portrait satisfies that route.
- Review the prompt, model, references, duration, output shape and quote. For cloned speech followed by video, review the speech step and its audio preview before the video step.
- Approve the intended work, then ask for its job status and output link. Accepted or queued is not finished.
- Play the completed clip, check identity, pronunciation and framing, then assemble and export in Editor if you need titles, music or several shots.
Versions, retries and deletion
- An avatar update creates a new version. Existing generation receipts keep the version and reference checksums that were used; editing the avatar does not rewrite completed work.
- If a save response is uncertain, inspect the current avatar or prior request before repeating it. API integrations must reuse the same idempotency key for the same retry. A stale expected version requires a refresh and review.
- Deleting an avatar removes it from future discovery and selection. It does not cancel already submitted generation jobs, erase billing records or automatically delete every source file.
- Deleting source media is a separate action with its own checks. Keep any local originals you need, and review the exact item named in a deletion confirmation.
Prepare speech, then animate
The current guided Clone video workflow separates Script & voice, Video model and Review & generate. Preview approved speech before animation; speech preparation and video generation have separate quotes. HeyGen Avatar IV and the offered Kling Avatar routes are available in that workflow. InfiniteTalk is retired for new requests.
The ordinary browser Editor keeps its working timeline in that browser. Do not assume an assistant’s server-side editor commands automatically modify that local timeline. Use the returned media/export and verify the intended editing surface.