How to use the AI Image Generator Studio agent skill

What does this agent skill make the agent do?

The skill gives an agent access to the Beatra image service from inside Claude Code, Codex or OpenClaw. It covers product images, brand visuals, posters, social graphics, illustrations and photo variations.

This catalog entry is a reviewed pointer, not the working package. It contains no client code and performs no Beatra operation by itself. The package it points to includes three standard-library Python scripts that make network calls and store a credential.

• Works on hosted Beatra models, not on your own computer.
• Does not update itself: it never downloads code or changes its own files.
• Is meant for paid work. Use it only when you accept that prompts and reference images go to a third party.

Before you install: requirements

You need Python 3.10 or later. The scripts use only the standard library, so no packages are installed. You also need git, because the install steps fetch the pinned package from github.com.

You need a Beatra account with paid credits, and you must approve a browser sign-in to Beatra. That sign-in stores a bearer token on your device.

• Python 3.10+ (standard library only)
• git, to fetch the pinned package from github.com
• A Beatra account and paid credits
• Browser sign-in to Beatra, which stores a bearer token

How do you install the skill?

Install only after reading the security notes below. The catalog's pinned steps fetch a specific commit, check its tree ID, and check for symlinks and executable files before anything is copied. Run these commands in a terminal, after you have approved the fetch.

Step 1 clones the source and checks the tree ID. Stop if the last line does not print TREE-OK. Step 2 lists any symlink, executable or submodule entries and searches for the old self-updater entry points.

Step 3 copies the reviewed folder into your skills directory. The destination must not already exist. Step 4 runs the sign-in script, which opens a browser and stores a token, so ask before running it. Start a new agent session if your host only finds skills at startup.

If you prefer not to use git, the brief for this page also describes downloading the ZIP from the page and unzipping it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills. Whichever route you choose, review the files first.

How do you ask the agent to use it?

Ask for the result you want and say that it should run on Beatra and that you accept the cost. The agent should first use free, non-billable discovery to read current model details and credit estimates. It then shows you a cost card for one paid call, listing the tool, model, estimate and any files it will upload.

Approval covers only what you named. Wait for an explicit yes before any paid render. If the request is open-ended, the agent should ask again rather than assume it can spend credits.

• "Using Beatra, make three square social graphics for our spring sale from the attached logo. Show me the cost before rendering."
• "Create a poster concept for a coffee shop launch on Beatra. Show me the estimate first and wait for my approval."
• "Check that the image skill can connect to Beatra. Do not run any paid job."

Check the connection and how a paid call works

The connection check is free and does not bill. Run it after authorizing to confirm the scripts can reach the service.

For a paid call, the agent uploads only files you supplied for this task and submits once with a stable request id. If a response is lost, it retries only with the same request id and identical arguments. It should never submit a replacement while a task is still running. When the result arrives, the agent reports the artifact and the net charge in credits and asks before any further paid step.

• Check with the verify command before any paid work.
• Upload only files you approve for this task.
• Review each result before asking for the next paid stage.

What should you check for safety?

This skill runs scripts, uses the network and stores a credential. Prompts and any reference images are sent to Beatra, which is a third party. Local-only editing is not supported because every render is hosted.

The scripts write local state, including credential and session files, and the catalog notes a version-notice behaviour. Read every file in the pinned folder yourself, including the scripts and references. Grep and manifest checks only help you find things; they are not a complete review. If you cannot review the code, do not install it and ask for a separate review instead.

• Runs scripts: yes. Uses network: yes.
• Stores a Beatra bearer token after browser sign-in.
• Use the same request id only when retrying the same paid call.
• To update, repeat the install steps for a new commit in a new review directory and approve the changes first.

How do you uninstall it?

Run the dry run first and read the JSON decision it prints. A result of disconnected means this was the last Beatra skill on the device. In that case the token is revoked when the output says so, and the local credential and state files are removed.

Run the real command only after the dry run looks correct.

• Files covered include credentials.json, installation.json, host.json, skills.json, network.json, balance-holds.json, pending_authorization.json and .authorize.lock.