One RTX 4080, One Night: Full AI Short Film With Qwen Images and Minimax H3 Video
CupQuakeBE · reddit · 2026-10-03
A creator tested the limits of local AI video production on a single RTX 4080 and finished an entire pipeline—from storyboard to 4K final frame—in one night, concluding commercial video models are no longer necessary for him.
His stack:
- Idea, full storyboard, location and monster design: human brain
- Prompt helper: Qwen 3.8 27B
- Images: Qwen image 2.1
- Video: Minimax H3 (single render with the Minimax H3 Director custom node for continuity)
- Upscaling to 4K: DLSS 5
The author says this is only a small part of a larger project that he expected to take months of trial and error, yet this demo scene—random detail shots pulled from his full storyboard—was incredibly fast and unexpectedly convincing on the very first try. He argues commercial models struggle to understand exactly what he wants, making the local setup feel like a cheat code, and even suspects commercial models introduce errors on purpose to burn credits. The whole run, from starting point to the last 4K upscaled frame, took one night on his normal computer.
More from Infra
- Chip design is a ~10^2,632,341 search problem — AI and agents are turning hardware into search — ai · 2026-10-03
- Redis Creator antirez Ships DwarfStar 4, an MIT-Licensed Local Inference Engine — petrusenko_max · 2026-10-03
- The $899 Mac Mini M6 was built for 24/7 agentic computing — most owners use it to check email — hey_abusiddik · 2026-10-03
- Modal VM Sandboxes hit GA: demo runs Docker Compose apps, tests and coding agents — charles_irl · 2026-10-03
- Community poll of 793 MLX users: oMLX wins at 55.5%, dominating Ultra chips — HankYeomans · 2026-10-03
- Local AI user asks: why does everyone enjoy taming the beast of complexity? — kathi7 · 2026-10-03