Exploration · Local AI video factory

HLN Machine

HLN Machine explores a local AI video factory. It is not about one generated result, but about splitting generation into a white-box pipeline that can be inspected, rerun, and repaired.

AI PipelineLocal-firstMedia AutomationWhite-box Engineering

Narrative

Design and engineering judgement

Product frame

Short-video generation looks like media automation, but the real work is workflow design: each intermediate state has to be understandable for the system to keep improving.

Engineering frame

The public Python pipeline coordinates news scripting, speech, A/B-roll, word-level subtitles, branding, and optional YouTube upload. Design notes, module specs, and runbooks make intermediate state and failure location inspectable.

Design frame

This is not a turnkey model or content bundle. It is a public AI engineering lab: the code and architecture are readable, while credentials, model weights, private media, and machine-specific assets remain operator supplied.

System

Components and visibility

Component Type Visibility Role
News story pipeline Editorial / speech stage Public snapshot Fetches source material, builds editorial context, drafts the script, synthesizes speech, and generates subtitle timing.
Video pipeline ComfyUI media stage Public snapshot Selects B-roll references and coordinates ComfyUI image, video, and speech-to-video workflows.
Branding pipeline Post-production stage Public snapshot Renders word-level subtitles and applies brand assets.
Upload pipeline Distribution stage Public snapshot Prepares title, description, and cover, with an optional YouTube Data API upload step.

Public URLs

Public surfaces