简体中文 · English

Lindyson AI Video Generator

Turn one product photo into publish-ready short videos and product images

An AI video and image generation workspace for e-commerce and short-video teams.
Credits, subscription plans, an admin console and an MCP endpoint are built in. Rebrand it and launch your own AI generation service.

Try it · Features · Keys · Deploy · FAQ · Development · Contributing · Security · Contact

Note

The product UI and the in-repo docs other than this file are in Simplified Chinese. The code, environment variables and commands are the same in both languages.

---

💡 Why We Open-Sourced This

We are cross-border sellers on TikTok ourselves, and a lot of the time we run our business on AI-generated video. Compared with shooting, it saves a lot of money and, honestly, it is easier to make something go viral. At seller meetups, more and more people came to us asking whether we could make videos for them too.

Over time we realized our own script writers and video people simply could not keep up with that much demand. That also told us the market is real.

So we built this system and are opening it up for everyone to use. The key part is the prompt system we refined on our own business: a customer uploads one photo, the system produces three prompt plans, and submitting one as-is already gives a very good result.

Those prompt plans are not set in stone. You are free to strengthen or rewrite them around your own content style, the skills you have accumulated, or your own logic, so the plans it produces fit your categories and your playbook.

Who it helps:

• Sellers running their own stores: upload a product photo, get three finished videos in different styles within about ten minutes, publish to TikTok.
• Agencies and MCN teams: batch generation across accounts and products, assets archived in one place, costs tracked in credits.
• Developers building AI generation products: a complete skeleton with billing, admin and an MCP endpoint, so you do not start from zero.

Commercial use is welcome. The code is Apache-2.0. Put your own brand on it and go live; you do not need to tell us and there is no revenue share. A few directions we think are worth doing:

• Private deployment for your own team or clients, on your own Volcengine account, with fully controlled costs.
• Plug in other multimodal models. The provider layer is isolated, so changing one place under apps/api/src/providers/ lets you swap in or run alongside models such as ChatGPT's Image 2.5 or MiniMax's H3, and let users pick by quality and price inside the same workspace.
• Connect your own payment channel and run it as a SaaS; plans, credits, orders and invoices are already there.
• Pull the prompt engine out on its own and integrate it into the sourcing or operations tools you already have.

The only requirement is not to use the "Lindyson" name and marks, see TRADEMARK.md. If you build something interesting, come back and tell us in an issue.

And if you are also a cross-border seller or in the TikTok business, whether or not you use this project, we would love to talk: product selection, making AI videos that go viral, which markets are good right now. We deal with these every day. Contact details are at the end under Contact.

---

🌱 Our Vision

We hope that one day a small seller, with no photographer and no editing team, can make videos as good-looking as a big brand's from a single product photo.

We hope this system does not just run in our hands. Everyone who deploys it can serve the sellers around them well, so more good products get seen by more people.

We also hope it keeps growing: someone plugs in a new model, someone rewrites the prompts to understand their category better, someone finds a use we never thought of, and brings that back so the next person has an easier start.

Doing business should not be a game only big companies get to play. A little more open tooling means a little more opportunity.

---

📸 Screenshots

Website

Publish preview	Pricing

User side

Dashboard	AI Studio: video generation
Digital assets	TikTok accounts
Publish to TikTok	Account center: credits and MCP connection

Prompt engine (core feature)

TK Prompt Engine starts from one product photo: it extracts product facts for you to confirm, lets you set the target market, voice-over language and visual approach, then generates three execution-ready storyboard prompts, "safest", "most native" and "most eye-catching". Pick one and it is written back to AI Studio.

Step 1: confirm product facts	Step 2: market and creative settings	Step 3: three prompt plans

Admin console

---

🚀 Try It in Five Minutes

Tip

No Volcengine account and no API keys needed. Demo mode swaps the real models for a "fake provider" that returns a placeholder about 1.5 seconds after each generation request, so the whole flow (upload, generate, assets, credits, admin) works end to end.

You need: a machine with Docker Desktop (or Docker Engine with Compose v2.24 or newer) and at least 4 GB of free memory.

Step 1: clone

git clone https://github.com/orylite/lindyson-ai.git && cd lindyson-ai

Step 2: one command to start

docker compose --env-file .env.demo -f docker-compose.yml -f docker-compose.demo.yml up -d --build

The first run builds the images, which takes about 3 to 5 minutes. Afterwards docker compose ps should show five running containers, postgres, redis, api, worker and web, with migrate exited (exit 0).

Step 3: open the browser

Go to http://localhost:8080, click "登录" (Sign in) and use the demo admin:

Email	Password
admin@example.com	Demo12345

Open AI Studio, upload any image and click generate. The task completes in about two seconds and a placeholder result shows up under digital assets. Then have a look at the admin console: users, orders, tickets and provider usage.

Stop and clean up

docker compose --env-file .env.demo -f docker-compose.yml -f docker-compose.demo.yml down

Add --volumes at the end to delete the data as well.

Note

The demo secrets and account are public, and the server refuses to start with these values in production mode, so they cannot be misused for a real deployment. To generate real videos, see Deploy to Production.

---

✨ Features

• 🎬 AI Studio: text-to-video, image-to-video, reference asset reuse, product image generation and interactive editing. The prompt wizard turns product information into storyboard scripts.
• 🧠 Prompt engine: extracts facts from a product photo and generates three execution-ready storyboard prompts for the target market and voice-over language.
• 🗂️ Digital assets: uploads, folders, thumbnails and batch operations, with images and videos managed together.
• 💳 Credits and subscriptions: plans, add-on packs, credit ledger, orders and invoices. Payment channels are configured in the admin console.
• 🛠️ Admin console: users, orders, tickets, announcements, payment channels, provider usage and cost reports.
• 🔌 MCP: a standard MCP server with OAuth 2.1 dynamic registration and PKCE, so ChatGPT and Codex can import assets and start generations directly. See MCP_CHATGPT.md.
• 📱 TikTok publishing: account management and publish preview; push finished videos straight to TikTok.
• 🧑 Real portrait library: consent, liveness check and revocation flows for real-person likenesses, with sensitive data stored encrypted.

---

🔑 Accounts and Keys You Need

The demo needs no keys. To generate real content and serve users, prepare the three groups below. Everything goes into .env; the full reference is .env.example.

Note

Generation is built on Volcengine Ark, ByteDance's model service platform. Seedance is its video generation model, Seedream is its image generation and editing model, and TOS is Volcengine's object storage.

Required: nothing real gets generated without these

What	Where to get it	Variable
Ark API key	Sign in to the Volcengine console, open Ark, create a key under API Key management, then enable these models under model activation: Seedance 2.0 (including fast and mini), Seedance 2.5, Seedream 5.0 Pro, and Doubao Seed 2.0 lite (used by the prompt engine). Pay-as-you-go; a small top-up is enough to start.	ARK_API_KEY
Object storage	Create a private bucket in Volcengine TOS, then create an IAM sub-user with read/write access to that bucket only and take its access key. Any S3-compatible storage also works.	STORAGE_DRIVER=tos and TOS_*

Tip

You are not locked into Volcengine. Generation goes through an isolated provider layer (apps/api/src/providers/) with one interface for video and one for images. To add ChatGPT's Image models, Google's Nano Banana or any other mainstream image or video model, write one provider following the existing Ark implementation and offer it to users alongside.

Optional by feature: leave it empty and the feature is off

Feature	What	Variable
Prompt engine	Reuses the Ark key above by default; it works as soon as Doubao Seed 2.0 lite is enabled. Create a separate key if you want separate billing. This layer is swappable too: DeepSeek or any other multimodal model can write the plans, and a different model with a tweaked strategy may give you something new to play with	ARK_PROMPT_WIZARD_API_KEY
Real portrait library	A Volcengine IAM sub-user access key for the Ark Assets API. Cannot reuse the Ark API key	REAL_PORTRAIT_PROVIDER_MODE=volcengine and VOLC_ARK_ASSET_*
Cost reports	A read-only access key for the Volcengine billing center, used to reconcile bills and resource packs in the admin console	VOLC_BILLING_*
Transactional email	A Resend account with a verified sending domain	RESEND_*
ChatGPT / Codex access	No external account; your site just needs a public HTTPS address	MCP_*, steps in MCP_CHATGPT.md
Baidu URL submission	A push token from Baidu Webmaster Tools	BAIDU_*

Generate yourself: five random secrets

These are not issued by anyone; generate them locally, one different value each, never reused:

Variable	Purpose	How
JWT_SECRET	Signs login sessions	openssl rand -hex 32
MCP_OAUTH_SECRET	Signs MCP OAuth tokens	openssl rand -hex 32
MCP_FILE_ENCRYPTION_KEY	Encrypts files imported through MCP	openssl rand -hex 32
REAL_PORTRAIT_TOKEN_ENCRYPTION_KEY	Encrypts real portrait credentials	openssl rand -hex 32
PAYMENT_CONFIG_ENCRYPTION_KEY	Encrypts payment channel secrets at rest	openssl rand -base64 32 (must be 32 bytes, Base64)

Payments: configured in the admin console, provider is your choice

Payment channel secrets are not in .env. Sign in to the admin console, open the payment channels page and enter the merchant number (mchNo), app ID (appId) and API key (apiKey) your provider gives you. They are encrypted with PAYMENT_CONFIG_ENCRYPTION_KEY before being stored.

Alipay and WeChat Pay can both be connected directly. The built-in channel is Jeepay: set up your own Alipay and WeChat Pay merchant accounts inside Jeepay, enter its merchant number, app ID and API key in the admin console, and customers pay through Alipay or WeChat directly. Which merchant accounts to apply for and under which entity is up to you. To use a different payment provider, the payment layer under apps/api/src/payments/ is isolated; add a channel following the Jeepay implementation.

---

💻 Local Development

Prerequisites: Node.js 22 or newer and Docker. You can run the whole flow without a Volcengine account: .env.example ships with ENABLE_FAKE_PROVIDER=true, so the worker produces placeholder results.

npm i

cp .env.example .env

Fill in .env following the comments in .env.example. Minimal local setup: set JWT_SECRET to any long random string, STORAGE_DRIVER=local, ENABLE_FAKE_PROVIDER=true, and leave the rest empty.

Start PostgreSQL and Redis (or point DATABASE_URL and REDIS_URL at your own instances):

docker compose up -d postgres redis

Initialize the database, seed the plans and create the admin (email and password come from BOOTSTRAP_ADMIN_EMAIL and BOOTSTRAP_ADMIN_PASSWORD):

npm run db:migrate && npm run db:seed && npm run db:bootstrap-admin

Start the API, the worker and the front end in three terminals:

npm run dev:api

npm run dev:worker

npm run dev

The front end runs at http://localhost:5173 and the API at http://localhost:4300; the dev server proxies /api/ to the API. You should see the landing page, and the dashboard after signing in.

---

🌐 Deploy to Production

A 2-core, 4 GB machine is enough: PostgreSQL, Redis, the API, the worker and the nginx front end all run from one docker compose.

Step 1: server and domain

A Linux server that can run Docker, and a domain pointing at it (ai.example.com in the examples below).

Step 2: clone and create the config

git clone https://github.com/orylite/lindyson-ai.git && cd lindyson-ai

cp .env.example .env

Step 3: edit .env, at minimum:

Variable	Value
NODE_ENV	production
JWT_SECRET, MCP_OAUTH_SECRET, MCP_FILE_ENCRYPTION_KEY, REAL_PORTRAIT_TOKEN_ENCRYPTION_KEY, PAYMENT_CONFIG_ENCRYPTION_KEY	Generate one each as in the section above; never reuse
POSTGRES_PASSWORD	Database password
PUBLIC_BASE_URL, APP_BASE_URL, CORS_ORIGIN	All https://ai.example.com
VITE_PUBLIC_SITE_URL	Same, used by the front-end build
ENABLE_FAKE_PROVIDER	false
ARK_API_KEY	Ark API key
STORAGE_DRIVER, TOS_*	tos and your object storage details
BOOTSTRAP_ADMIN_EMAIL, BOOTSTRAP_ADMIN_PASSWORD	The first admin account

Step 4: operator and compliance details

The footer and legal copy read src/app/data/publicCompliance.json; the one in the repo is a placeholder template. Copy it to publicCompliance.local.json in the same directory (git-ignored) and fill in your real company name, registration numbers and privacy contact. The build swaps it in automatically. npm run check:public-compliance verifies it is complete.

Step 5: start

docker compose up -d --build

docker compose up runs database migrations automatically. After the first start, seed the plans and create the admin:

docker compose run --rm --no-deps migrate npm --workspace apps/api run db:seed:prod

docker compose run --rm --no-deps migrate npm --workspace apps/api run db:bootstrap-admin:prod

Opening http://your-server-ip in a browser should now show the landing page.

Step 6: add HTTPS

The easiest option is Caddy in front, which obtains certificates automatically. First change WEB_HTTP_PORT in .env to an internal port (for example 8080) and TRUST_PROXY to 2 (two hops: Caddy and web), restart with docker compose up -d, then write a Caddyfile:

ai.example.com {
    reverse_proxy 127.0.0.1:8080
}

Start Caddy, open https://ai.example.com, sign in as admin, configure payment channels, plans and announcements in the console, and you are open for business.

Warning

The API should only listen on localhost; public traffic must come through the reverse proxy. Set TRUST_PROXY to the proxy's address, CIDR or hop count. Do not set it to true while the API port is directly reachable, or clients can spoof their source IP and bypass rate limits.

Deploying without Docker

The API and the worker are the same codebase. After npm run build:api, start them with node dist/server.js and node dist/worker/generationWorker.js; migrate with node dist/db/migrate.js; seed and create the admin with node dist/db/seed.js and node dist/db/bootstrapAdmin.js. For the front end, run npm run build and serve dist/ from any static server, proxying /api/, /mcp, /oauth/ and /.well-known/oauth-* to the API; see infra/nginx/default.conf.

Other configuration notes

• Front-end build-time variables (VITE_ prefix, can live in .env.local): VITE_PUBLIC_SITE_URL is the public site URL used for canonical links and social sharing; VITE_SUPPORT_QR_URL is the support QR image URL, hidden when empty. VITE_SEEDREAM_TUTORIAL_MEDIA_BASE_URL points at the directory holding the Seedream tutorial media, which are official Volcengine assets not distributed with the repo; when empty the dialog shows text and official links only.
• Change the domains in public/robots.txt and public/sitemap.xml to your own.
• Production mode refuses known weak default secrets, http site URLs and public storage URLs. If startup fails, fix what the error names.

---

🧯 FAQ, Backup and Upgrade

The page is a 502 or blank after startup

Check docker compose ps first. If migrate did not exit 0, read docker compose logs migrate; it is usually a wrong database password or DATABASE_URL. If api keeps restarting, docker compose logs api prints which environment variable is missing.

Creating the admin fails with "an active pro base plan must exist first"

The database has no plan data yet. Run the seed once (npm run db:seed locally, or docker compose run --rm --no-deps migrate npm --workspace apps/api run db:seed:prod in Docker), then create the admin. The seed is idempotent and never overwrites existing data.

Generation tasks stay queued with no result

The worker is not running or cannot reach Redis. Check docker compose logs worker; in local development the worker needs its own terminal running npm run dev:worker. In real mode also confirm ARK_API_KEY is valid and the models are enabled.

What does the fake provider produce

A placeholder image or video. Task IDs start with fake- and nothing is charged on Volcengine. It only works when NODE_ENV is not production.

Port 80 or 8080 is already in use

Change WEB_HTTP_PORT in .env (or .env.demo) and run up -d again.

Backup: data lives in two Docker volumes, postgres-data (the database) and local-storage (files when using local storage). One-line database backup:

docker compose exec postgres pg_dump -U lindyson lindyson > lindyson-$(date +%F).sql

With object storage, files are in your TOS bucket; back them up with your cloud provider's tools.

Upgrade:

git pull && docker compose up -d --build

up applies new database migrations automatically. Back up before upgrading.

---

🧩 Development Guide

Architecture

flowchart LR
    B[Browser] -->|static files| W[web<br/>nginx + Vite/React build]
    B -->|/api/ · /mcp · /oauth| W
    W -->|reverse proxy| A[api<br/>Fastify]
    A --> P[(PostgreSQL 16)]
    A --> R[(Redis 7 · BullMQ)]
    R --> K[worker]
    K -->|Seedance / Seedream| V[Volcengine Ark]
    K -->|files| S[(Object storage<br/>TOS or local disk)]
    A --> S
    C[ChatGPT / Codex] -->|MCP + OAuth 2.1| W

Layout

Directory	Contents
src/	Front end. index.html → src/landing.tsx is the landing page; app.html → src/main.tsx is the signed-in workspace and admin console
apps/api/	API, worker, database migrations (migrations/), operational tools (src/tools/)
apps/api/src/providers/	Provider layer. Change this to swap models or add vendors
packages/api-contracts/	zod contracts shared by front end and API
tests/	Playwright end-to-end tests and Docker full-stack tests
scripts/	Full-stack test orchestration, public compliance validation, Baidu URL submission
docker-compose.yml	Single-machine production stack; docker-compose.demo.yml + .env.demo is the demo overlay; docker-compose.fullstack.yml is the test overlay

Commands

Command	What it does
npm run typecheck / npm run typecheck:api	Type-check the front end / the API
npm run test:unit	Front-end and contracts unit tests (Vitest)
npm run check:api	API type check and unit tests
npm run test:e2e	Playwright end-to-end tests against a mock API
npm run test:fullstack	Brings up real PostgreSQL, Redis, API, worker and browser with docker-compose.yml plus docker-compose.fullstack.yml
npm run build / npm run build:api	Production builds; the front-end build also emits .gz for text assets
npm run check	All of the above; run before a release

Contribution flow is in CONTRIBUTING.md, security issues go privately through SECURITY.md, and changes are tracked in CHANGELOG.md.

---

📮 Contact

If something breaks, open an issue. To talk about commercial use, private deployment, plugging in other models, or just to swap notes on TikTok short video, scan the Feishu (Lark) code:

_{Scan with Feishu / Lark}

---

⭐ Star History

If this is useful, a star helps more sellers and developers find it.

---

📜 Third-Party Assets and License

Third-party assets distributed with the repository are listed in NOTICE, with details in ATTRIBUTIONS.md and public/fonts/说明.md. The product images and videos on the landing page are AI-generated and documented in src/imports/landing-masters/PROVENANCE.md; the banner at the top of this README is a screenshot of the landing page. Master files are not in the repository.

This project is released under the Apache-2.0 license. The "Lindyson" name and marks are not licensed with the code; services built on this project should use their own brand, see TRADEMARK.md.