Define the audience. Shape the brief. Direct your social video.
Define your audience and plan detailed single-avatar social videos in ChatGPT or Claude, with platform-specific briefs, camera timing, performance, captions, sound and matching HeyGen text.
Social Media Avatar AI Director is an audience-first planning package for ChatGPT and Claude. It helps educators, creators, subject experts, small businesses and content teams turn a social-video idea into an audience profile, a focused brief and detailed instructions for one selected HeyGen avatar. The creator reviews the plan, copies the matching text into HeyGen and handles generation and finishing manually.
Its central value is a clearer, more reviewable production brief: audience assumptions stay distinct from supplied evidence, social decisions are explicit, and spoken words stay separate from production notes and posting text. Time savings, fewer generation attempts, revenue gains and audience performance have not been measured or guaranteed.
These are intended audiences based on package scope, not evidence of validated demand. Product stories need real, supplied support; hypothetical examples are not proof.
The workflow helps identify a primary viewer’s needs, knowledge, viewing context and desired takeaway. It can compare up to three labeled hypotheses and distinguish creator-reported or hypothetical claims from evidence-linked ones. Choosing or reviewing a profile does not validate demand, prove a segment exists, or establish that a message will perform.
After selecting a primary audience, the creator and agent shape the platform and placement, organic or paid context, objective, one message, truthful opening hook, delivered payoff, sharing reason and a useful next action—or no call to action when none is appropriate. The brief can support explainers, practical tips, problem/solution, myth correction, checklists, product stories based on supplied assets, insights and one-person micro-stories.
A platform is not an audience, and a general social brief does not establish platform eligibility or performance. Cross-posting may need a separate adaptation of the hook, layout, caption, cover line, action and measurement. Paid-creative guidance is not evidence of organic results.
Cinematic detail is optional. The package does not require rapid cuts or treat an example timing template as a universal attention rule.
Outputs are separated for the confirmed workflow: a project prompt where appropriate, literal spoken script, supported performance directions, timed planning details and a distinct posting package. The posting title or cover concept, post caption and optional relevant hashtags are not text to paste into a spoken-script field. A cover concept is not a generated image.
Account-specific controls and route compatibility must be confirmed by the creator. Missing essentials remain questions or a draft rather than being presented as verified.
Social Avatar Workspace v2.0 replaces the two v1 studios with one standalone HTML workspace. Quick start, deeper fields, editable defaults, readable draft handoffs and local import/export help organize a brief. It does not call an AI model, generate media or connect to an external service.
Preserve creator purpose, brand voice, supported claims and assets alongside audience hypotheses and availability evidence. Compare substantive angles, assess avatar suitability and review the CTA destination and fulfillment. Confirmation records a planning choice; it does not establish market demand.
Prepare separate outputs for Video Agent, Presenter/AI Studio and Quick Avatar. Body and face motion stay separate from camera and scene direction. Exact-word locks, desired beat timings and optional declared pilot measurements support literal-text timing calculations. Planning declarations, actual-media review and the manual results ledger remain separate; affected changes revoke relevant confirmations.
Optional capacity-aware series records retain original metric definitions and comparison guidance. Alternatively, work directly in ChatGPT or Claude with the portable instructions and reference knowledge. Local browser checks reported by the package do not establish behavior in hosted previews or successful customer host setup.
Breaking v1 migration: the format is social-avatar-workspace-2.0. Only video-audience-1.0 profiles have an explicit draft importer; mapped values need review and evidence is not promoted. Old scene/social/bundle JSON is not automatically migrated. Avatar Shots/Single Scene short-source routes are not implemented in the v2 compiler. Keep original files and consult the included migration guide.
The package prepares text and planning guidance. It does not operate HeyGen, submit prompts, spend credits, inspect generated footage, edit video, mix audio, publish content or retrieve performance data. Desired timing is a production target: preview voice, inspect footage and allow for further editing. Exact cuts, acting, avatar consistency, lip-sync and runtime are not guaranteed.
The workspace assembles requests and preserves editable choices; it contains no AI model. The package has its own skill and workspace and does not depend on or change Movie Scene Director or another director skill.
For ChatGPT or Claude Projects, the guide describes adding the portable instructions and reference files. The inner installer is an optional native Claude Skills route where the account or organization supports it. Actual account controls and host setup can differ; successful installation or native loading is not established by the supplied package.
You need your own ChatGPT or Claude access, suitable HeyGen access, an eligible avatar and voice option, and any media assets and permissions needed for your work. The creator supplies the audience context, exact words, proof and creative requirements, verifies available account controls and reviews any generated result. When sharing information with those services, their processing is governed by those services; a local artifact does not make the overall workflow offline.
Creators remain responsible for factual review, proof, asset rights, platform and placement checks, permissions, generation decisions and manual finishing. No commercial licence terms, support arrangement or future-update promise are established here.
The supplied v2.0.0 package reports 60 logic/CLI checks, 17 JSDOM interaction checks and 9 standalone headless Chromium checks—86 total, plus five fresh-context text tasks used during iteration. These are package-reported checks, not independently verified by MiReVerse.
Browser scope covered the local HTML file, download/import, a narrow viewport and absence of HTTP requests during the tested flow. Selected text responses show sampled planning behavior, not statistical validation or guaranteed model behavior. Examples and ready fixtures are fictional or simulated. Automated checks do not verify factual prose, rights, disclosure judgments or self-reported media review.
Native Claude installation/loading, ChatGPT or Claude Project setup, novice usability, hosted previews, account-specific HeyGen controls, generated footage and audio, publication acceptance and marketing effectiveness remain unverified. No accounts were connected and no generation credits were spent. Desired seconds, captions, acting, cuts and other generated requests require creator review.