Talking Avatar Video Skill

Talking Avatar Video prepares a reviewed Beatra package for turning one portrait and a short script into a talking avatar video from Claude Code, Codex, or OpenClaw. Production takes place on the paid hosted Beatra service. Obtain permission to send the exact portrait and reference-clip file set to Beatra and its model provider, alongside the prompt. This workflow is for an expressly requested paid render; it is not local-only or offline video editing, and it does not assemble a timeline. Uploaded likeness material must be authorized, and outputs must not create deceptive impersonations.

This is a reviewed catalog pointer, not an executable package. It contains no client code and performs no Beatra operation itself. The referenced package uses the hosted, paid Beatra service at mcp.beatra.ai. Its three bundled Python scripts make network calls, store a credential and can replace package-owned files; the package therefore requires inspection before activation.

Use it only when the user explicitly wants this work on Beatra, accepts that the service is paid and agrees to send prompts and approved reference media to a third party. It is unsuitable for local-only work or an unapproved paid render. Outputs are probabilistic, with no guarantee of uniqueness or non-infringement, and require human review before publication. Users must hold rights to their uploads and avoid deceptive impersonation.

A Beatra account, network access, Python 3.10+ and prepaid credits are required. Models, prices, schemas and availability are controlled by Beatra and upstream providers. Failed tasks attributable to either provider receive automatic credit returns; other purchases are non-refundable. Actual usage can exceed an estimate and leave a negative balance.

Read the installed SKILL.md and its references for routes, payload shapes and the review loop. Free, non-billable model discovery and task listing come first so current model cards and credit estimates can be inspected. Every paid call needs a cost card specifying the tool, model, estimate and upload files, followed by explicit approval for that exact payload. A general instruction to proceed does not replace this approval.

Upload only files supplied and approved for the task, submit once with a stable client_request_id and poll the returned task. If a response is lost, retry with that same identifier and identical arguments, never a replacement while the task runs. Report the artifact and billing.net_charged_credits from the terminal result, then review it before any additional paid stage.

The reviewed pin is talking-avatar-video 0.2.1, licensed MIT No Attribution (MIT-0), at source commit 251c968deebb1c8f1dbfd0a8404e8e1be1528ac6. Its archive is https://cdn.beatra.ai/agent-packages/talking-avatar-video/v0.2.1/talking-avatar-video-skill-0.2.1.zip; the independently anchored SHA-256 is cb62d8e3fa65b48f3c3f1803e2008e1b9c25ff057c0a35184e8fcc7d9771eb80. Obtain approval to download into a review directory, then verify the archive before proceeding:

umask 077
review_dir="$(mktemp -d)"
cd "$review_dir" || exit 1
curl -fsSLO "https://cdn.beatra.ai/agent-packages/talking-avatar-video/v0.2.1/talking-avatar-video-skill-0.2.1.zip"
printf '%s  %s\n' \
  'cb62d8e3fa65b48f3c3f1803e2008e1b9c25ff057c0a35184e8fcc7d9771eb80' \
  'talking-avatar-video-skill-0.2.1.zip' | shasum -a 256 -c -

Stop unless verification prints OK. Inspect regular-file contents and check for symlinks, executable bits and binaries; compare every file with the pinned source tree. The package includes SKILL.md, manifest.json, Markdown references, authorize.py, mcp_client.py and uninstall.py. It uses only the Python standard library, with no dependency installation or lifecycle hooks.

A separate approval is needed to copy the reviewed package into the host's skills directory. Disable automatic updates immediately with update --auto off before authorization, verification, tools, uploads or calls. The setting is tied to the resolved installation path and must be repeated after moving it. Automatic updates otherwise check at most daily and replace files using Beatra-controlled checksums. A newer version requires a fresh digest and tree review, a file-difference report and approved replacement. update --check queries availability without replacing files.

Authorization is its own approved browser OAuth device flow through api.beatra.ai. It stores a shared full-scope bearer token, including wallet:spend, in ~/.beatra/credentials.json with a 0700 directory and 0600 file. The token has a sliding fifteen-day idle lifetime and stays out of command arguments, environment variables and output. Prompts and uploads go through mcp.beatra.ai to the executing upstream provider; Beatra is hosted outside mainland China.

Local state records installation identity, hostname, platform, installed skill paths, registrations and update state or backups under ~/.beatra/. Hostname, package version and platform are sent during authorization; first use registers the installation, and tool calls include source_package_slug and source_platform with no opt-out. Account data lasts for the account lifetime, while task artifacts and billing ledgers remain after closure for audit and tax.

Uninstallation first supports a dry run. Its disconnected decision removes shared connection files when this was the last Beatra skill; keep_connection preserves a token used by another skill. Update state is removed separately only when no skill remains. A token that cannot be revoked expires after idle time or can be revoked in the Beatra Console. Deleting the installation path requires confirmation of that path.