Reddit user makes two videos in 6 hours with a fully local AI video stack

Unlucky_Principle480 · reddit · 2026-11-16

A Reddit user shared two low-budget videos made for the Hermes Plugin Hackathon with a fully local stack: OpenMontage for orchestration, gpt-image keyframes with Opus 5.5, LTX-2.5 image-to-video in ComfyUI, HyperFrames for UI/text, Qwen3-TTS and Fish Audio Drama 3 voice clones, and Stable Audio 3 for music. Both videos took about 6 hours including planning and rendering. They flag LTX-2.5's artifacts (cat physics issues), weak instruction following, and limited realism, asking for better local image-realism and image-to-video models.

Original post →

More from coding & agent

coding & agent channel →