Marketing Studio

One-click, self-hostable AI e-commerce ad SaaS, powered by the AtlasCloud API. Ship a credit-metered video-ad product on Cloudflare Workers in minutes.

Four production-grade pipelines, each wiring multiple models into one flow:

App	What it does	Pipeline
UGC Product Ad (/marketing-studio)	Product + presenter photos → lip-synced UGC ad	AI prompt-expand → first frame (nano-banana/edit) → seedance-2.0 i2v
Reference to Ad (/ad-reference)	Upload a viral ad → remake it with your product & presenter	gemini-omni-flash/video-edit swaps product+talent → optional ElevenLabs TTS + veed/lipsync; auto-fallback to kling motion-control for real-person talent
AI Drama Ad (/drama-studio)	One topic → comedy script → shot-by-shot drama ad	LLM script → cast/scene/product reference images → seedance-2.0/reference-to-video per shot
Ad Skit (/ad-skit)	One-line product → two-hander comedy skit	LLM script → gpt-image-2 product shot → seedance-2.0/reference-to-video (15s, with audio)

All four auto-detect the input language — write your topic in any language, get the script/ad in that language.

Tech stack

• Next.js 14 (App Router) + OpenNext on Cloudflare Workers
• Prisma on Cloudflare D1
• AtlasCloud API for every generation step (300+ image/video/LLM models behind one API)
• Credit metering with dynamic per-second × resolution video billing (below)
• NextAuth (Google) login · Stripe or redeem-code top-ups

Quick start

git clone https://github.com/AtlasCloudAI/atlas-ecommerce-studio.git
cd atlas-ecommerce-studio
cp .env.example .env      # fill ATLASCLOUD_API_KEY + DATABASE_URL
npm install
npm run dev               # http://localhost:3000

Get a free AtlasCloud key at atlascloud.ai.

Deploy to Cloudflare

npm run cf:deploy         # prisma generate + opennext build + wrangler deploy

Set secrets (ATLASCLOUD_API_KEY, NEXTAUTH_SECRET, …) via wrangler secret put or the Cloudflare dashboard — they never live in the repo.

Credits & pricing

Users spend in-app credits; you pay AtlasCloud in USD. The mapping is cost-based:

• Fixed-cost steps (images, LLM scripts, TTS) → flat credits.

• Video is billed dynamically — src/lib/video-pricing.ts computes credits from each model's real per-second × resolution rate:

  credits = ⌈ perSecond[resolution] × seconds × ACCOUNT_MARKUP × MARGIN / CREDIT_USD ⌉
  

  e.g. seedance 720p = $0.242/s, gemini-omni-flash/video-edit = $0.14/s, kling motion-control = $0.112/s. 720p/short → fewer credits; 1080p/long → scales up. Rates verified against production billing; tune MARGIN in one place. Top-up packs live in src/config/pricing.ts.

License

MIT. Built on AtlasCloud.