Let your AI build your twin and videos

Connect this app to my AutoPersonify account so you can turn people or characters into realistic video twins and make presenter videos with them.

## How this works
- You (the AI) help me build a twin of a person or character, then use that twin to make many videos over time.
- Every video you make is a draft until I, the human, watch it and approve it in this app. Never publish or place a video on the site automatically.

## What AutoPersonify is
AutoPersonify (https://autopersonify.com) is a UK AI video platform. It makes finished presenter videos from a script, using either a stock AI presenter or the user's own "digital twin", which is a realistic video likeness of a person or character with a voice. Accounts are billed by minutes of finished video. Videos made through the API use the account's normal plan allowance and also appear in the user's Projects. It cannot build a person from a written description alone: a twin needs at least one real photo (or a short video).

## Docs
- Human docs: https://autopersonify.com/developers.md
- OpenAPI spec: https://autopersonify.com/openapi.json
- Base URL: https://autopersonify.com/api/v1. Every request must send the header Authorization: Bearer <key>. Requests and responses are JSON.

## Endpoints
- GET /me: plan, video seconds used and remaining, twins_used, twins_allowed, can_use_twin_in_videos, upgrade_url
- GET /presenters: stock presenters
- GET /twins, GET /twins/{id}, POST /twins, DELETE /twins/{id}: the user's twins
- POST /uploads: returns a private upload_url for a photo (image/jpeg, png or webp, up to 50MB) or a video (video/mp4 or video/quicktime, up to 500MB). Always upload the original full-resolution files — never resize or compress them; the twin is built from the original file exactly as sent.
- POST /videos, GET /videos/{id}: make a video and check on it
- GET /voices: standard voices plus the user's own starred voices, with gender and preview details where available

## Do this, in order
1. Ask me for my AutoPersonify API key. I create it in AutoPersonify → Integrations → Developer API & AI agents → Step 2. It starts with ap_live_. Store it as a server-side secret named AUTOPERSONIFY_API_KEY. Never expose it to the browser or a mobile app. Make every call from server code.
2. Call GET /me and tell me my plan, the video minutes I have left, how many twins I've used out of my allowance, and whether my twins can be used in videos. If I'm short on any of these, show me the upgrade_url.
3. Build a webhook receiver: a public POST route in this app, for example /api/webhooks/autopersonify.
   - Read the raw request body.
   - Verify the X-AutoPersonify-Signature header. It must equal "sha256=" followed by HMAC-SHA256(secret, X-AutoPersonify-Timestamp + "." + raw body), as hex. Compare in constant time, and reject timestamps older than 5 minutes.
   - Handle these events: twin.awaiting_approval, twin.ready, twin.failed, twin.rejected, video.ready, video.failed.
   - IMPORTANT — order of operations: the signing secret does not exist yet. Do NOT ask me for AUTOPERSONIFY_WEBHOOK_SECRET at this stage. First give me the full public https URL of the webhook route and tell me: "Paste this into AutoPersonify → Integrations → Developer API & AI agents → Step 3 (Webhook address), then give me the signing secret it shows you."
   - Only after I paste the signing secret back to you, store it as AUTOPERSONIFY_WEBHOOK_SECRET. Until the webhook is set up, poll instead.
4. To build a twin:
   - First call GET /voices. For a photo twin, either show or play suitable choices and ask me to confirm the exact voice, or send voice_personality: a short description of the narrator I want (gender, accent, warmth, pace — for example "warm, authoritative British male narrator, calm pace"). AutoPersonify matches the closest catalogue voice from that description and I still see and approve the choice on the approval card. Keep the chosen voice_id; never silently guess.
   - For each photo, call POST /uploads {content_type}, then PUT the file to the upload_url it returns.
   - Then call POST /twins with { name, context, photos: [upload_ids], voice_id or voice_personality, likeness: { type: "original_character" | "real_person_consented", consent_by, statement } }. Photo twins need voice_id (exact id from GET /voices) or voice_personality (a short description of the narrator). Put the sharpest, evenly lit, front-facing photo with a neutral expression FIRST. Always include context: one sentence saying what the twin is for and which video you plan to make first (for example "Welcome video for the homepage"), so the owner can see what they are approving.
   - Optionally add one training video, passing its upload_id as "video". About 2 minutes of the person talking to camera gives the most realistic result.
   - Nothing happens until I trigger this step (for example a button in this app), because it uses my twin allowance.
   - The twin starts as "awaiting_approval" and I get an email. Tell me to open AutoPersonify → Your twin and tap Approve.
   - Show the twin's live status in this app (awaiting approval → building → ready / failed / rejected), updated from the webhooks, or by polling GET /twins/{id} every 30 seconds. Building usually takes a few minutes after approval. AutoPersonify only shows progress inside AutoPersonify, so this app must show its own status.
   - Before offering to make a video, check video seconds remaining from GET /me. Free plans include 3 minutes that work for any video, including twin videos. If no time is left, show the upgrade_url instead of calling POST /videos.
5. To make a video:
   - Write the script yourself, using what you know about my app, brand and audience.
   - Call POST /videos { twin_id or presenter_id, script, title, idempotency_key }. It returns 202 with status "rendering" and an id. Nothing is final until you see "ready".
   - While the video is being made, never leave me guessing. In this app show a clear, live production status wherever the video will appear (for example in the Twin Video Library and on the page that will use it): a status label plus live progress percentage from GET /videos/{id} (use its progress and human-readable stage fields — the same figures shown in the AutoPersonify dashboard), for example "Rendering — 45%…", alongside the title, a "Made with AutoPersonify" line, the script preview and how long it has been rendering (e.g. "1m 40s so far — usually 2–10 minutes"). Poll GET /videos/{id} every 10–15 seconds, or use the video.ready / video.failed webhooks. Refresh the status automatically; do not require me to reload the page.
   - On failure (video.failed or a failed poll), switch the status to "Failed", show the error's user_message, keep the video in the library with a Retry action, and tell me what went wrong — do not silently hide it.
   - When it is ready, immediately switch the status to "Needs review", show the video in the library player (play/pause, scrubbing, mute, full screen) ready for me to watch, and tell me it is ready to review — the first thing I should see is the finished video, not a form or a link.
   - Save the video id, not the link, because links expire after an hour. Fetch a fresh link from GET /videos/{id} whenever the video is played or thumbnailed.
6. Build a Twin Video Library page in this app (admin/owner only):
   - A dedicated admin page named after the twin (for example "Max Twin"), placed next to the app's brand or media settings. If the app already has a page for the twin's imagery, move it into this page.
   - Twin section: name, photo, voice and live status (awaiting approval, building, ready, failed).
   - Video library: one row or card per video with thumbnail, title, script, length, created date and status (Rendering with a live percentage while it is being made, Ready for review, Approved, Rejected, Live, Failed). Keep every version.
   - A proper player: play/pause, scrubbing, mute/volume and a full-screen view, with a poster thumbnail. While a video is rendering, show the progress status in its place; when it is ready, the player takes over immediately.
   - Actions: Approve, Reject (with an optional note), and Request changes (you write a new script and make a new version).
   - Only Approved videos can be placed on the site (for example the homepage hero). Show where each video is used.
   - Fetch a fresh link from GET /videos/{id} whenever a video is played. Optionally copy approved files into this app's own storage for permanent hosting.
   - Update statuses from the video.ready and video.failed webhooks, or by polling.
7. Errors come back as { error: { code, message, user_message, action, upgrade_url } }. Always show me user_message. If action is "upgrade", show an Upgrade button linking to upgrade_url. Retry 503s with backoff (5s, 20s, 60s), at most 3 times. Don't retry 401 or 402.

Example request: "Make Max (our AI agent) into a real-looking person with AutoPersonify's twin product using the photos in /public/max, then build a 30-second welcome video for the homepage and put it in the Max Twin page for my approval."