VibeSKU Clips

中文版 | English

VibeSKU Clips turns product links, photos, and a brief into a downloadable 15-second UGC product video. Choose a product, performer, scene, format, language, and market; review the script, optionally review a storyboard, then confirm video generation. Each work stays in one list from creation through download.

The platform produces video assets. Storefront links, inventory, publishing, and performance tracking stay with the team that posts them.

What it does

• Product material and facts. Import a page through Firecrawl or upload up to six product photos. Extracted facts become usable when parsing succeeds and remain editable on the product page. Warnings are advisory; missing essential material blocks generation.
• Reusable talent and scenes. Describe a fictional adult performer or supply authorized reference photos. Talent and scene libraries each generate one square reference sheet that can be reused across works.
• Eleven formats. Presenter, everyday scene, tutorial, tech demo, beauty routine, food and drink, apparel, styling, fit check, accessories, and unboxing each have their own script and shot vocabulary.
• Two generation modes. One-take draws an opening frame before generating video. Storyboard mode draws one frame per beat in sequence and waits for approval. Scripts always stop for review; storyboard and video generation require confirmation.
• Language and market controls. Clip language, target market, and interface language are separate. The interface supports English and Simplified Chinese; clip languages include English, Spanish, Portuguese, Japanese, Korean, and Simplified Chinese.
• Versions and downloads. Review generated versions, revise the script or frames, and regenerate. Video and SRT subtitles are archived in private R2 storage; scripts retain publish captions and AI disclosure text.
• Durable tasks. PostgreSQL, pg-boss, and a task-run outbox preserve progress through restarts. Failed or stalled steps surface in the console with retry actions. Usage records retain generation costs and output lineage.

Every clip targets 15 seconds with 9:16 or 16:9 framing. Models and resolutions depend on the selected provider:

Video provider	Model	Resolutions
Prism	H3	480p, 720p
lk888	H3	720p, 1080p, 2K
lk888	Seedance 2.0/2.5	480p, 720p, 1080p

The current quality report checks script lengths and records the requested clip duration. It does not independently measure the video duration or inspect the rendered product, identity, and language. Review the actual video before use.

Production flow

The default path is product → script review → video → download. Storyboard mode adds a reviewed storyboard between script and video. When a product is still being read, the work waits; once usable, it starts the script automatically.

Job	Purpose
ugc.product.ingest	Import page material and extract product facts
ugc.talent.generate	Generate a reusable performer reference sheet
ugc.scene.generate	Generate a reusable scene reference sheet
ugc.work.script	Write a script and wait for review
ugc.work.storyboard	Draw sequential key frames and wait for approval
ugc.work.video	Generate video, archive media, and record quality checks

The Web process enqueues tasks. The separate Node Worker calls Firecrawl, the LLM, and media providers. Prism generates images for both video backends. Private image references are signed at the Worker boundary, and provider outputs are copied to R2 because their original URLs expire.

Stack

Next.js 16 App Router, React 19, TypeScript, Tailwind CSS v4, shadcn/ui, next-intl, PostgreSQL, Drizzle ORM, pg-boss, Better Auth, Resend, Cloudflare R2, Stripe, and Vercel AI SDK v7. Repository Markdown is compiled with Content Collections.

Quick start

Requires Node.js 22.12.0 or newer, pnpm 10.33.0, and PostgreSQL. Rendering also requires your own LLM, image/video provider, and R2 credentials; these hosted services are separate from the source-code license.

git clone https://github.com/UllrAI/VibeSKU-Clips.git
cd VibeSKU-Clips
pnpm install --frozen-lockfile
cp .env.example .env
openssl rand -base64 32

Use the generated value for BETTER_AUTH_SECRET, then fill in .env with your own service credentials. Placeholder values do not make a working installation.

Configuration

.env.example is the configuration template. Web validates its settings in env.js; the Worker validates its subset in worker-env.ts.

Settings	Purpose
DATABASE_URL	Required PostgreSQL application database
JOB_DATABASE_URL	Optional separate queue database; defaults to DATABASE_URL
NEXT_PUBLIC_APP_URL, BETTER_AUTH_SECRET	Public origin and a random session secret of at least 32 characters
RESEND_API_KEY, RESEND_EMAIL_FROM	Magic-link authentication; sender must use a verified domain
LLM_API_KEY, LLM_BASE_URL, AI_DEFAULT_MODEL	LLM access; defaults to OpenRouter and openai/gpt-5.6-luna
FIRECRAWL_API_KEY, FIRECRAWL_API_BASE_URL	Worker product URL imports; default root is https://api.firecrawl.dev/v2
PRISM_API_KEY, PRISM_API_SECRET, PRISM_API_BASE_URL	Worker image generation and Prism video; credentials must match the selected host
VIDEO_GENERATION_PROVIDER	prism (default) or lk888; configure the same value for Web and Worker
LK888_API_KEY, LK888_API_BASE_URL	Required key for lk888 video; default root is https://api.lk888.ai
R2_ENDPOINT, R2_ACCESS_KEY_ID, R2_SECRET_ACCESS_KEY, R2_BUCKET_NAME	Private uploads and generated media, shared by Web and Worker
STRIPE_SECRET_KEY, STRIPE_WEBHOOK_SECRET, STRIPE_ENVIRONMENT	Billing; matching test/live credentials and webhook signature secret
GOOGLE_*, GITHUB_*, LINKEDIN_*	Optional OAuth client ID/secret pairs
DB_POOL_SIZE, JOB_DB_POOL_SIZE, WORKER_GRACEFUL_TIMEOUT_MS	Connection budget and Worker shutdown timeout
RATE_LIMIT_IP_HEADER	Trusted ingress header; defaults to x-forwarded-for; the proxy must overwrite client input
BING_SITE_VERIFICATION, NEXT_PUBLIC_UMAMI_*	Optional deployment-specific site verification and analytics

SITE_CONFIG in src/lib/config/site.js owns brand, contacts, links, and the emailAuth, billing, uploads, and ai switches. All switches default to enabled. Disable unused features before removing their credentials. With email authentication disabled, configure at least one complete OAuth provider. The AI switch controls the assistant; clip production still needs its Worker integrations.

Prism defaults to staging in development and production in production builds. Set PRISM_API_BASE_URL explicitly for your account. The development Docker Compose example defaults to staging even though its containers use production builds.

Keep the R2 bucket private and allow your application origin in its upload CORS policy. Direct uploads need PUT, Content-Type, and If-None-Match; see Docker setup and architecture.

Example R2 CORS policy; replace the origins with your own:

[
  {
    "AllowedOrigins": ["http://localhost:3000", "https://yourdomain.com"],
    "AllowedMethods": ["PUT"],
    "AllowedHeaders": ["Content-Type", "If-None-Match"],
    "ExposeHeaders": ["ETag"],
    "MaxAgeSeconds": 3000
  }
]

Database and development

Create the database referenced by DATABASE_URL, then apply committed migrations:

pnpm db:migrate
pnpm dev

Open http://localhost:3000. pnpm dev runs both Web and Worker with file watching. pnpm dev:web starts only Web; queued jobs need a separately running Worker.

The Worker also applies committed migrations at boot under a PostgreSQL advisory lock, including pg-boss setup. Repeated boots are safe; migration failure stops the Worker. Web never migrates. The Worker database role therefore needs schema migration permissions. Use pnpm db:generate for shared schema changes; reserve pnpm db:push for disposable local iteration.

Sign up normally, then grant admin access explicitly if needed:

pnpm set:admin --email=your-email@example.com

Optional billing setup

With billing enabled, configure your own Stripe account and run:

pnpm stripe:sync-products

This updates the selected test/live catalog in src/lib/billing/stripe/prices.ts. Existing catalog IDs belong to the reference deployment; they are public identifiers, but another Stripe account cannot reuse them. Review the generated diff before deployment. See webhook operations.

API and CLI authentication

Browser users use session cookies. Machine endpoints under /api/v1/* use user-managed API keys or browser-approved CLI sessions. Manage both at /dashboard/developer.

pnpm vibesku-clips-cli -- auth login --base-url http://localhost:3000
pnpm vibesku-clips-cli -- auth status --base-url http://localhost:3000

Scripts may use VIBESKU_CLIPS_CLI_API_KEY instead. The current CLI supports login, status, refresh, and logout; it is an authentication tool. See CLI documentation.

Development checks

Command	Purpose
pnpm lint	ESLint
pnpm type-check	Generate content/route types and check TypeScript
pnpm test --runInBand	Jest tests
pnpm test:e2e	Build and run Playwright browser tests
pnpm test:jobs:integration	Durable-job database integration tests
pnpm dead-code:check	Unused files, exports, and dependencies
pnpm prettier:check	Formatting
pnpm build	Standalone Web build plus Worker artifact
pnpm worker:start	Run the built Worker
pnpm analyze	Production bundle analysis

E2E requires E2E_DATABASE_URL pointing to a dedicated local database whose name contains a standalone e2e or test segment. Never use a regular application database. The runner applies migrations and cleans its fixtures; it generates a per-run test secret when one is not provided.

Deployment

Run both Web and Worker from the same release. pnpm build prepares the standalone Web output and dist/worker/worker.mjs. pnpm start runs only Web; run the Worker separately with pnpm worker:start and supply its environment through your process manager. NEXT_PUBLIC_APP_URL must be correct at build time. Use /api/health for liveness and /api/ready for database readiness.

• Docker setup: a development Compose stack with PostgreSQL, migrations, Web, and Worker. Example database credentials are for local use.
• Zeabur deployment: separate Web/Worker services, production migration access, backup configuration, and fork instructions.

The reference deployment tracks prod. A reviewed default-branch commit is promoted by pushing a release/vX.Y.Z tag matching package.json. The release workflow checks Quality for that exact SHA and applies migrations before moving prod; do not push directly to that branch. Forks configure their own GitHub production secrets, migration tunnel, and branch filters.

Repository and self-hosting notes

• Replace brand, support/legal/privacy contacts, social links, and blog author information for your deployment. Credentials never belong in client-safe SITE_CONFIG or NEXT_PUBLIC_* variables.
• Use your own Stripe catalog, OAuth applications, verified email domain, R2 bucket, analytics website ID, and Bing verification token.
• The code contains Prism production/staging API roots. Source access does not provide service access; configure an authorized provider account.
• Keep .env files, private keys, database backups, logs, customer material, and generated production media out of Git. .env.example contains placeholders.
• Open-source review records the current repository inspection and outstanding dependency/configuration concerns.

See AGENTS.md for repository conventions, design.md for UI guidance, background jobs for task operations, and AI agents for the assistant architecture. Report vulnerabilities privately through SECURITY.md.

License

MIT. Dependency licenses and third-party service terms remain separate. The homepage images are AI-generated concept photography; provenance is documented in landing assets.