Use an LLM as the orchestrator
Built a Python pipeline in which an LLM orchestrates script generation and coordinates specialised subagents for the production workflow, using LangChain and Ollama / vLLM.
CASE STUDY / GENERATIVE AI · VIDEO PRODUCTION PIPELINE
From a script to an assembled AI video
An end-to-end video pipeline orchestrating scripts, speech, image and video synthesis, and final assembly with LLMs and specialised subagents.
01 / THE PROBLEM
What needed to work.
Producing an AI-generated video requires more than generating a clip. Scripts, speech, visual assets, and final assembly need to work together. This project connects those stages in an end-to-end LLM-orchestrated pipeline.
02 / MY APPROACH
How I approached it.
Generate speech and visual material
Connected KokoroTTS with image and video synthesis stages, using stable diffusion and Wan2.2 as part of the generation toolchain.
Assemble the final output
Integrated FFmpeg for final video assembly, bringing generated material together into production-ready videos.
03 / SYSTEM OVERVIEW
Connecting the pieces.
- 01Script generation
- 02Subagent orchestration
- 03Speech + image/video synthesis
- 04FFmpeg assembly
- 05Finished video
The core focus was connecting specialised generation tools into a complete workflow. The orchestration layer coordinates creative stages, while the assembly stage turns individual generated assets into a finished output.
AI VIDEO / PRODUCTION PIPELINE03
01Script
02Voice + visuals
03Assembly
One orchestrator. A complete creative workflow.
04 / OUTCOMES
What the system delivers.
Connected workflow
Script generation through final video assembly.
Watch the output
Example outputs are available on the linked YouTube channel.
Independent project overview. No production throughput, latency, or cost benchmarks have been supplied.