Launching a Multi-Model AI Content Platform
A subscription-based SaaS platform for AI image and video generation, built end-to-end with React 19, NestJS, AWS, and 15 integrated AI models
Business Need
Voxii Labs had a clear vision for a subscription-based AI content platform but no technical foundation to build it on. The company needed a development partner capable of owning the full product lifecycle: from infrastructure design and backend architecture to frontend delivery and AI provider integration. Without an end-to-end team, the product could not reach the market.
Result
Todor3D delivered a fully operational SaaS platform in 4 months, meeting every milestone in the agreed delivery plan. Voxii.ai launched with 15 integrated AI models across image and video generation, a Stripe-powered subscription system, and an active base of paying tier 3 subscribers.
From Vision to Blueprint
Voxii Labs came to Todor3D with a product idea and a concrete market opportunity: a platform where subscribers could generate AI images and video on demand using best-in-class models. What they did not have was any technical foundation. We opened the engagement with a structured discovery phase, working directly with the client to define the full product scope, map the system architecture, and produce a delivery plan with concrete milestones and agreed dates.
The output of discovery was a full technical blueprint covering three distinct layers: the main user-facing website, a dedicated admin panel, and a backend API. Infrastructure was designed around AWS Ubuntu instances with GitLab CI/CD pipelines for reliable, automated deployments. We introduced Cloudflare R2 for scalable media storage alongside AWS S3 and AWS MediaConvert for video processing, establishing the backbone before a single feature was built.
Key items
- Full discovery phase with scope definition and milestone planning
- Three-layer architecture: main website, admin panel, backend API
- AWS Ubuntu + GitLab CI/CD infrastructure from day one
- Cloudflare R2 and AWS S3/MediaConvert for media storage and processing
One Team, Full-Stack Ownership
With the architecture confirmed, our team of five took end-to-end ownership of every layer of the product. The backend API was built on NestJS 10 with TypeScript, using PostgreSQL managed via TypeORM and a JWT and Passport authentication layer. Stripe handled subscription billing, SendGrid covered transactional email, and media processing ran through FFmpeg and Sharp. The entire API was documented with Swagger throughout development to ensure maintainability after handoff.
The main website was built in React 19 with Vite, Tailwind CSS, Framer Motion for interface animations, Video.js for media playback, and Sentry for error monitoring. Progressive Web App support was included, giving users a native-like experience across devices. The admin panel used a parallel React 19 and Vite stack with Radix UI, Zustand for state management, and React Query for data fetching. This gave the Voxii team full operational control over users, subscriptions, and content without relying on the development team for day-to-day management.
Key items
- Backend: NestJS 10, PostgreSQL, TypeORM, Stripe, SendGrid, FFmpeg, Sharp, Winston, Jest
- Main website: React 19, Vite, Tailwind CSS, Framer Motion, Video.js, Sentry, PWA
- Admin panel: React 19, Vite, Radix UI, Zustand, React Query, SASS
- Swagger API documentation included throughout
15 Models, One Unified Interface
The defining capability of Voxii.ai is its AI module: a unified interface that gives subscribers access to leading generation models without managing API keys, rate limits, or infrastructure on their own. We built a consistent abstraction layer so that every model, regardless of provider, is exposed through the same interface on both the frontend and the admin panel.
For image generation, we integrated six models: Runway ML Gen4 Image, Kling Image v3, Gemini Nano Banana, Flux.1 Kontext, Seedream 4.5, and Qwen Image. For video generation, we integrated nine models: Kling v2.5 Turbo, Seedance 1.0 Pro Fast, Seedance 2.0, Seedance 2.0 Fast, Kling v3, RunwayML Gen 4.5, Veo 3.1, Sora 2, and Wan 2.2. The platform launched on schedule. Every milestone in the delivery plan was met. Voxii Labs went live with an active subscriber base and paying tier 3 users within the agreed 4-month timeline.
Key items
- Image generation: Runway ML Gen4 Image, Kling Image v3, Gemini Nano Banana, Flux.1 Kontext, Seedream 4.5, Qwen Image
- Video generation: Kling v2.5 Turbo, Seedance 1.0 Pro Fast, Seedance 2.0, Seedance 2.0 Fast, Kling v3, RunwayML Gen 4.5, Veo 3.1, Sora 2, Wan 2.2
- Unified API abstraction layer across all providers
- Subscription tiers enforced at the API level via Stripe integration
15
AI models integrated (6 image + 9 video)
100%
milestones delivered on schedule
Gallery
Build a High-Conversion
3D Experience for Your Brand
Let's design a configurator that increases engagement and drives measurable growth.
Recent Experiments
Exploring the future of interactive commerce and WebGL experiences.
UAE Energy Distribution Utility
Labrador
Clash of Diggers
Simplio
Thermogreen
