Open-source h3pipe pipeline ships a 20+ minute AI-generated series episode
bgstratt · reddit · 2026-10-05
A developer released h3pipe, an open-source AI video pipeline (github.com/bgstratt/h3pipe), and used it to produce the first 20+ minute episode of an AI series, Porchlights S1E01.
Workflow:
- A written script (packaged as a skill) plus series. feeds a shotlist. that drives the pipeline, conforming prompts to each model's style
- Multi-model video support (Wan, LTX, H3), reference image generation (Krea, Qwen, Z-Image, Flux, H3), and sound generation
- Keybind shortcuts to mark shots for review, reprompt, and slicing
Lessons from long-form AI video:
- Keeping background extras consistent across shots and maintaining spatial integrity in tight rooms are the hardest problems; the author fell back to generating room "tours" with H3 and snapshotting frames
- Putting all characters in the same scene keeps conversations grounded and prevents characters popping in and out
- Most work was a loop of review, rewrite, references, and reprompting
The author invites community tips on extras consistency, FL2V, and background plates.
More from coding & agent
- GLM vs Qwen tool calling differences force separate handlers in Codex proxy — TheZachMueller · 2026-10-05
- Why running third-party models in Codex needs a tool-calling proxy layer — TheZachMueller · 2026-10-05
- Dev says nearly all his web searches now run through personal agents as web goes headless — illscience · 2026-10-05
- RelayDesk connects Claude to your own computers and VMs via MCP — whateverxp · 2026-10-05
- Musk amplifies "zero human company" run by a Grok bot as CEO — elonmusk · 2026-10-05
- The AI Checkout Demo Everyone Wants: Agent Retries Without Double-Charging — HaktanSuren · 2026-10-05