This article was first published on my site: https://resume.tensorview.cc/blog/ai-video-creation-engine.html
FastAPI + GPU / NVENC: business logic decoupled from render
I built a distributed AI video splitting and synthesis backend. FastAPI handles logic on CPU nodes while encode/decode rides NVIDIA GPUs via FFmpeg CUDA (NVENC) — about 8–10× faster than CPU-only rendering.
Mini-program “AI Original”: split, scenes, voiceover, one-tap synthesize.
Key features
- Heterogeneous compute: logic decoupled from GPU render for concurrent jobs.
- AI workflow: DeepSeek rewrite plus Edge-TTS voiceover.
- Storage: local cache + Qiniu OSS with LRU cleanup.
- Pro codecs: lossless 4K splitting and real-time multi-track mix.
AI script UI: DeepSeek create/rewrite with Edge-TTS voice pick.
History: synthesis status, preview, and download.
Closing
Whether you run a short-video matrix or automated content pipelines, this stack is built for faster synthesis. Always happy to talk optimization or A/V automation.
Original post: https://resume.tensorview.cc/blog/ai-video-creation-engine.html
Top comments (0)