# Salk AI

> Salk AI ("Automate Attention") is a canvas-based AI studio for making video ads. Users describe what they want or pick a viral-ready template; a chat-driven Claude agent ("Salk") drops sticky notes, queues AI generations (image, video, voiceover, lipsync), and runs local audio edits onto a spatial workspace. Pay-as-you-go — every major image/video/voice model in one place, no subscription.

## What Salk does

Salk turns a one-line brief into a finished video ad. A single project mixes:

- **Image generation** — multiple text-to-image providers
- **Video generation** — Google Veo (3.1 / 3.1 Fast / 3.1 Lite / 3 / 3 Fast / 2), Kuaishou Kling (v1 → v3 Omni, including `std` / `pro` / `master` tiers), and image-to-video flows
- **Talking-head avatars** — HeyGen Avatar III / IV / Video Agent, including user-trained Photo Avatars and a synced stock catalogue
- **Voiceover** — ElevenLabs text-to-speech (eleven_v3, multilingual_v2, flash_v2_5, turbo_v2_5) and speech-to-speech, with a curated multi-language voice catalogue
- **Lipsync, audio trim/speed, and other local FFmpeg edits** in the browser
- **A/B variant fan-out** — one prompt with variables spins out dozens of ad variants in parallel
- **Templates** — a registry of "viral-ready" template JSONs the user fills in with their product, logo or brand voice
- **Reel matching** — paste an Instagram / TikTok / YouTube reel URL and Salk picks the closest template

## Who it's for

- Founders shipping creative on day one
- Performance marketers ironing out new variants weekly
- Creators turning ideas into polished reels
- Small agencies delivering more ads to more clients without scaling headcount

## How it works

1. **Describe your ad** — type a one-line brief, pick a template, or paste a reel URL
2. **Salk builds it** — drafts the script, generates visuals, records a voiceover, stitches it together
3. **Tweak and ship** — fine-tune on the canvas, spin out A/B variants, export ready-to-post video

The workspace is a spatial canvas — *not* a chat window. Elements (images, videos, audio, sticky notes, runs, avatars, pipelines) connect via typed ports (image=orange, video=purple, audio=green, text=blue, avatar=pink) so creators can wire shots, voiceovers and edits together visually.

## Stack at a glance

- **Frontend** — Next.js 15 (App Router), React 19, TypeScript strict, Tailwind (dark-only "plum-noir" theme)
- **Auth & multi-tenancy** — Clerk (email + Google), Prisma-backed `User` / `Organization` / `OrgMembership` mirror; active org lives in the `salk_active_org` httpOnly cookie
- **Data** — PostgreSQL via Prisma 7 (driver adapter `@prisma/adapter-pg`)
- **Object storage** — Google Cloud Storage (bucket `ai-videos-platform`); all asset URLs are proxied through `/api/assets/{assetId}`
- **Agent** — Anthropic Claude SDK; the Salk agent speaks in user-facing terms only ("image", "video", "voiceover", "clip", "edit"), never "model", "tool", "skill", or "id"
- **Wallet** — pay-as-you-go credits (`WalletTransaction` / `WalletPurchase`), Razorpay for top-ups; new orgs start with 500 credits

## Public surfaces

- [Landing](/) — product overview and template gallery
- [Pricing](/pricing) — credit packs and pay-as-you-go pricing
- [Done For You](/done-for-you) — managed-service offering for teams who want Salk to run their ad creative
- [Privacy](/privacy)
- [Terms](/terms)
- [Login](/login) / [Signup](/signup)

## Contact

- Email: hello@salk.ai
- Tagline: *Automate Attention.* Made for creators who'd rather ship than wait.