Description

Bicep Heart AI is a specialized video generation platform that transforms a single photograph into a short, dynamic video featuring the signature “bicep heart” pose — a social media trend where one arm flexes while the other hand curves into a heart shape beside the bicep. Built for creators, fitness enthusiasts, and anyone looking to join the viral bicep heart movement, the platform delivers studio-quality pose animations from just one uploaded image.

The workflow is designed for clarity and control. Users upload a photo, review the exact credit cost before committing, then generate their video. No hidden charges, no subscription traps — just transparent, pay-as-you-go credit packs. Each completed video can be previewed in the browser and downloaded for sharing.

Under the hood, the application runs on a modern, production-grade stack that balances developer experience with deployment flexibility. TanStack Start provides the full-stack React foundation with file-based routing and server functions. Drizzle ORM handles database operations across PostgreSQL, MySQL, SQLite, and Cloudflare D1, making it trivial to move from local development to cloud deployment. Better-auth manages authentication flows — email/password, OAuth with Google and GitHub, password reset, and email verification — all backed by a granular role-based access control system that ships with super admin, admin, and member roles out of the box.

Internationalization is first-class, powered by Paraglide JS. The entire interface ships in English and Chinese, with a compile-time approach that tree-shakes unused translations. There’s no provider to mount and no hook to import — just compiled message functions that work everywhere, from components to server loaders to API routes.

The payment and credits architecture is battle-tested. Stripe, PayPal, and Creem integrations handle one-time purchases and subscriptions, with webhook handlers that atomically grant credits, create orders, and update subscription status. Credits use FIFO consumption, support expiration dates, and track every transaction for full audit trails. The admin panel gives operators complete visibility — user credit balances, transaction logs, manual grants for customer support, and fine-grained RBAC controls.

Storage sits behind a unified interface with adapters for AWS S3 and Cloudflare R2. Generated videos are uploaded to object storage — a hard requirement on Cloudflare Workers, where there’s no local disk — and served through signed URLs that expire after a configurable window. Reference photos use short-lived signed URLs sent directly to the video provider.

The AI layer is built to swap providers without rewriting call sites. The platform supports OpenAI-compatible endpoints targeting MiniMax H3 video generation, plus Tuzi’s MiniMax Video API as a fallback. Every provider configuration lives in the database and is encrypted at rest when CONFIG_ENCRYPTION_KEY is set. Operators test credentials through the admin panel before going live — each provider group has a “Test” button that makes a real request, so you catch a bad API key before a customer does.

The admin experience is comprehensive. Beyond settings and provider configuration, there’s full user management — search, filter, view credit balances, grant credits manually, assign roles. Invite codes can be generated in batches with optional expiration dates, usage limits, and trial periods. Subscriptions, orders, and credits are tracked in separate tabs with pagination and search, backed by TanStack Table with manual server-side pagination.

The architecture is module-first. Every business domain — payments, subscriptions, credits, RBAC, API keys, AI tasks — lives in its own service file under src/modules/, importing only from core infrastructure, config, or Drizzle. Modules don’t know about each other, so you can delete features you don’t need without chasing import graphs.

Deploying to Cloudflare Workers is streamlined through build scripts that handle nitro preset selection, D1 setup, migrations, and secret management. The entire stack runs on edge compute with millisecond cold starts. For teams that need more control or want Postgres at scale, the same build targets Node and runs on any host — Docker, VPS, or a managed container platform. The database abstraction means you change one environment variable and the schema template, then push or migrate as usual.

The codebase separates structure from content. Reusable components live in src/components/ and take all their content via props. Page sections — header, hero, features, pricing, footer — live in src/blocks/ and wire translations into those components. This split is what survives a rebrand: components are the durable chassis you keep, blocks are content you rewrite. The default landing page is demo material, expected to be replaced.

Bicep Heart AI is a turnkey solution for video generation with the business infrastructure to match. It’s not just a script that hits an API — it’s a full SaaS engine with payments, user accounts, RBAC, credits, multi-language support, and an admin panel. Whether you’re launching a fitness content tool, a social media trend platform, or a white-label video service, the foundation is already built.