Hermes Agent + LTX 2.3: ComfyUI story pipeline tutorial

This tutorial connects Hermes Agent to ComfyUI for a local AI video storytelling pipeline. The presenter uses a custom visual-storyline-pipeline skill to guide story development, image creation and video generation. ComfyUI workflows must first be exported as API-format JSON, and the ComfyUI backend must be running for generation.

The process begins with text: a storyline and character list come before visual assets. Z Image Turbo generates characters and locations, while reference images support keyframe creation. The presenter requests character sheets with front, side and back views plus a close-up, all on a white background. Review pauses let the user correct images before committing computing resources to video.

The LTX workflow includes OmniNFT, a LoRA that the speaker says improves motion and audio-video synchronization. The tutorial describes an FP32 download and a smaller BF16 option for ComfyUI. These are the presenter's claims, rather than a measured comparison in the supplied material.

The demonstration generates five scenes from a planned set of 30 and shows how conversational instructions can pause or revise the task. Human review remains necessary: the presenter points out unwanted eye and shirt colors, and says the resulting clips may need trimming or further editing because some actions are inaccurate.