The Problem

Creating documentary-style videos traditionally requires significant team resources and multi-day timelines. The challenge was compressing the entire production workflow into something a single person could run in minutes, without producing generic-quality AI output.

What I Built

3DLabs is an end-to-end system that transforms a topic into a finished, production-ready documentary video in roughly 8 minutes. The pipeline runs through ten sequential stages: script generation, scene planning, image creation, animation, voice narration, sound design, and multi-format export, coordinating over a dozen AI models through a unified orchestration layer.

Technical Highlights

Visual Coherence Managed prompts and seeds across multiple generation calls to maintain consistent imagery throughout the video. The hard part wasn't calling AI models, it was making the pipeline hold together. Model Orchestration Built a unified routing layer across Gemini, Replicate (Flux, Seedance, LTX, Minimax), and ElevenLabs, abstracting each provider's quirks behind a single interface. Cost Management Implemented a pre-flight credit reservation system to prevent runaway inference expenses across long, multi-stage generation runs. Multi-Tenancy Production-ready authentication and workspace isolation via Clerk, so multiple users can run pipelines concurrently without cross-contamination.

Stack

Next.js · TypeScript · Google Gemini · Replicate · ElevenLabs · FFmpeg

Result

• Finished, production-ready videos in ~8 minutes from a single topic input
• Eliminated days of manual production work per video
• Coordinated a dozen+ AI models through one orchestration layer without manual glue work