Documentation

AdneticAI turns a product, URL or brief into finished video ads: research, hooks, scripts, AI actors, generation, editing, localization and performance learning in one workflow.

Your first ten minutes

• Create an accountwith email (you'll confirm your address via a link) or Google. Then onboarding asks what you want to create, who you are, and where you advertise — this tunes defaults, it never locks features.
• Pick a plan and pay with PayPal. Every plan is credit-metered; credits are the single currency for AI work (see Credits).
• Create your first brand — name, website, logo. If you enter a website, the URL importer can extract product facts for you to confirm.
• Generate: the fastest path is URL → Ad: paste a product page, confirm the extracted facts, pick a hook, and queue your first video.

Workspace concepts

• Organization — your account's billing and team boundary. All credits, plans and members live here.
• Workspace — a container inside the organization (agencies use one per client).
• Brand — identity, assets, voice and rules. Every creative belongs to a brand.
• Campaign — a brief plus its hooks, scripts, creatives and approvals.

The Brand Kit (Creative Studio → Brand Assets) is the source of truth the AI works from: identity, visual assets, voice, and the claims it may — and may not — make.

• Assets: upload logos, music beds and other files. The logo powers branded end cards in the editor.
• Voice & audience: brand voice notes and target-audience context flow into every hook and script the AI writes.
• Approved / prohibited claims: the compliance reviewer blocks scripts that contradict these.
• Auto-localization languages: pick target locales and every approved creative is automatically queued for localization into them.

Creative Studio → Products holds the products you advertise. Products arrive two ways: imported from a URL (the importer extracts facts, features and proof for you to confirm) or created manually.

• Images: upload photos, logos, packaging shots and screenshots per product (JPEG/PNG/WebP/GIF, up to 25MB each).
• Facts, not fiction: the AI only uses claims supported by your source pages or your own input — unsupported claims are flagged, never invented.
• Shortcuts: every product card links straight into hook generation.

Paste a product URL and the research pipeline crawls the public page, summarizes the product, extracts features, benefits, proof points and objections, and proposes personas and creative angles. You confirm the extracted facts before anything is saved — nothing is stored or used until you approve it.

From the confirmed research you can generate ranked hooks, full scripts, and go straight to casting and video generation without leaving the flow.

The Hook studio generates opening lines across styles — curiosity, question, contrarian, problem, proof, transformation, listicle and more — and scores each for scroll-stop potential, clarity, relevance and compliance risk.

The Script studio writes complete scripts using proven frameworks (Hook → Problem → Solution → CTA, PAS, AIDA, Before/After/Bridge, testimonial, demo, myth-busting and others), with word counts and an estimated spoken duration. Every script passes the compliance reviewer: unsupported claims, risky health/financial statements and deceptive testimonials are flagged with reasons, and blockers stop generation.

The Actor Library offers curated presenters with photo previews (hover to see them talk). Filter by presentation, age band, language and style; favorite the ones you use often. Voices come from the voice catalog — click any voice to hear a preview before committing.

Custom actors & voice cloning

• Create a custom actor from your own footage, or clone a voice from a sample.
• Both require an explicit consent and rights attestation — you confirm you own or have licensed the likeness and voice, and the attestation is recorded with a timestamp. Non-consensual cloning is not supported, full stop.
• Custom actors and clones belong to your organization and are never shared.

Create supports several modes, all sharing the same pipeline:

• UGC Video — an AI actor delivers your script in a natural setting.
• Product Video — showcase scenes: reveals, unboxings, lifestyle shots, hero shots.
• App / SaaS Ad — actor-plus-screen formats for software.
• Talking Actor — a presenter speaks your script or uploaded audio.
• AI B-Roll — supporting footage from a scene description.
• Video Remix — new hook, captions, CTA or language on an existing creative.

Every generation shows a credit estimate before anything runs; you confirm, credits are reserved, and the job queues. If a job fails, the reservation is released — failed generations are never charged. The model router picks the best available provider for the request; you can also select a model explicitly. Video length follows your script's spoken duration, so narration is never compressed.

The editor assembles finished ads from generated clips. Pick clips from the bin, then work the timeline — a scrubber strip with filmstrip thumbnails, real audio waveforms and a playhead synced to the preview player.

• Cut: trim by dragging clip edges (or type exact seconds), split at the playhead, reorder, delete.
• Pace: per-clip speed 0.5×–2× with pitch preserved; per-clip mute; a master narration volume.
• Transitions: crossfade or fade-to-black between clips and into the end card.
• Text: up to 8 burned-in text/CTA overlays with position, size and timing.
• Audio: brand music bed with volume and auto-ducking under narration.
• Frame: output at 9:16, 1:1, 4:5 or 16:9, letterboxed or center-crop filled, 720p or 1080p (plan-dependent).
• Captions: word-timed captions in multiple styles, burned in or sidecar.

The preview honors trims and speed and plays the whole timeline through — what you see is what renders. Renders are priced by output length and shown as an estimate first.

The Creative Matrix multiplies dimensions — hooks × actors × CTAs × styles — and shows the full combination count. The AI ranks combinations, you select which to render, and you see the total credit cost before confirming. Nothing renders automatically.

Every variation stays linked to its parent, so you can trace any creative back to the exact hook, script and actor that produced it.

Localization is transcreation, not word-for-word translation: message, idioms, CTA and captions are culturally adapted, and a locale-compatible voice re-records the narration. Claims are preserved exactly — adaptation never changes what a product promises.

Localize on demand from any creative, or set auto-localization languages on the brand so every approved creative fans out automatically. Localized videos keep their link to the original.

Campaigns collect the brief, strategy, hooks, scripts and creatives in one place, moving through draft → internal review → client review → approved → exported.

• Review in editor: reviewers watch the creative in the editor and can fix issues before approving.
• Client review links: share a scoped, token-protected link — clients comment and approve without an account.
• Comments & decisions are logged with timestamps; a rejection moves the campaign to changes-requested.
• On approval, auto-localization (if configured) and webhooks fire.

Connect your ad account (Meta first) and import per-creative metrics: impressions, CTR, CPC, CPA, ROAS, video-hold rates and more, normalized across platforms.

• Creative Genome: every creative is described by its traits — hook type, angle, actor profile, pacing, CTA — so performance can be correlated to what the creative actually is.
• Fatigue detection: declining CTR and rising costs are flagged with confidence and evidence, never as certainty.
• Recommendations & Mutation Engine: keep what works (actor, offer, CTA), mutate what's tired (hook, first 3 seconds, captions). Suggested mutations never spend credits until you accept them.

Campaign Autopilot runs the whole chain — research, personas, strategy, hooks, scripts, compliance, casting — then stops and shows you the plan with a credit estimate. Nothing generates until you approve.

• Workflow canvas: compose reusable strategy workflows from the same skills; canvas runs produce artifacts (research, hooks, scripts) and never spend generation credits.
• Scheduled refresh: periodically re-generate creatives from winning traits, within a strict per-run cap and full compliance review.
• Fatigue-triggered suggestions: when a creative fatigues, a mutation suggestion appears — you decide whether to run it.

Credits meter all AI work. The flow is always: estimate → you confirm → reserve → run → capture actual cost. If the actual cost is lower than the reservation the difference is released; if a job fails the whole reservation is released.

• Plans (Starter, Creator, Growth, Agency, Business) grant monthly credits and features; billing is via PayPal, monthly or yearly.
• The full ledger — grants, reservations, captures, releases — is visible in Billing. Nothing is charged invisibly.
• Plan limits (seats, brands, resolution caps) are enforced at generation time with a clear message, not a silent failure.

Invite teammates by email from the Team page. Roles: owner (billing, deletion, everything), admin (manage content, team and settings), member (create and edit). Invitations are emailed and auto-accepted when the invitee signs up with the invited address. Agencies typically use one workspace per client plus client review links for approvals.

Everything the UI does is available programmatically. Start at the API reference.

• REST API v1 with scoped API keys (created in Developers; shown once, stored hashed).
• Webhooks for generation, creative, localization, workflow and credit events — HMAC-signed, replay-protected, retried with delivery logs.
• TypeScript SDK (@adnetic/sdk) with typed clients and a jobs.waitFor helper.
• MCP server at /api/mcp exposing the same services to AI agents.
• Enterprise SSO (OIDC/SAML per email domain) with organization provisioning.

AdneticAI generates realistic people and advertising claims, so safety is enforced server-side, not by UI convention:

• Custom actors and voice clones require recorded consent; revocation deletes the actor.
• The compliance reviewer checks every script and blocks deceptive testimonials, unsupported claims and restricted categories.
• Generated media carries provenance — model, prompt, inputs, seed — for every output.
• Synthetic-media labels are supported, and provider watermarks/credentials are never stripped.
• Tenant isolation is enforced on every query; media URLs are signed and short-lived.