One prompt, 8.5-minute sitcom: fully local video pipeline runs on a single RTX 4090
teachersecret · reddit · 2026-10-14
A Redditor demonstrated a fully automated local video generation pipeline: given a single prompt — "make a 1980s sitcom called Dorm Daze" — the system produced an 8.5-minute previz-style video over a few hours on a RTX 4090, with no human intervention.
The stack is entirely local and open:
- Orion-26B-A4B LLM (with mmproj), swapping between VRAM and DDR4, handles script, scene, camera work, and editing;
- Image/video generation via ComfyUI + Minimax H3 at 3 steps with a turbo LoRA for speed;
- Audio generated up front through his self-built minimax-h3-voice-api;
- The remaining automation (camera review and editing) is done by his modified Bantam Factory, which he jokingly calls the "chicken factory."
He acknowledges the camera work has issues, but it's a fun proof that a single prompt can drive the whole video production pipeline locally.
More from coding & agent
- Dev builds scope-locked local agent harness: GLM 5.3 Flash fixes a real bug PR-by-PR — HankYeomans · 2026-10-14
- Claude Code session startles dev with a familiar name — turns out it's MCP Agent Mail — doodlestein · 2026-10-14
- AIPOCH's open-source Open-Science workbench: reproducible AI research sessions — ahuja_priyank · 2026-10-14
- LSCBench: Vibe Coding Agents Write Functional Code, But Only 25% Is Also Secure — Qwen · 2026-10-14
- sshexport 2.9 ships with ChatGPT-audited hardened code and OpenWRT support — esrtweet · 2026-10-14
- AI agent deletes entire edited season of Swedish reality show weeks before premiere — torb · 2026-10-14