VideoForge

VideoForge is an end-to-end, production-grade AI video synthesis studio built to orchestrate state-of-the-art Diffusion Transformer (DiT) video models on consumer-tier workstation hardware (e.g., NVIDIA GeForce RTX 4060 with 8 GB VRAM). Instead of relying on costly third-party cloud APIs or requiring server-grade enterprise accelerators (24GB+), VideoForge leverages asymmetric dual-bus memory offloading, flow matching sigma schedules, and native spatio-temporal guidance (STG) to generate high-fidelity, temporal-coherent videos locally in ~30–45 seconds while strictly maintaining a peak VRAM footprint of ≤5.2 GB. The project follows Clean Architecture and Ports & Adapters (Hexagonal Architecture), strictly decoupling client-facing transport layers, task orchestration, hardware adapters, and local background inference engines. It supports four distinct generation modalities within a unified builder: Text-to-Video (T2V), Single Image-to-Video (I2V), Start→End Frame Transition, and Multi-Frame Keyframing (up to 4 custom keyframe anchors). The system includes a REST API, WebSocket push server for real-time telemetry, and a React 18 + Vite SPA frontend. Hardware benchmarks on an AMD Ryzen 7 8845HS, 64 GB DDR5, NVIDIA GeForce RTX 4060 Laptop (8 GB VRAM) show render times of 30.1s (Draft), 44.5s (Balanced), and 55.8s (Extended) with peak VRAM between 5,207 MB and 5,620 MB.