An AI-assisted, human-led video production pipeline. From idea research to published episode — structured, traceable, and open source.
Turn one approved video idea into one controlled, credible, traceable, and editable episode package.
The system automates production plumbing while preserving human creative authority. The objective is not maximum asset generation — it's making each episode worth producing, clearly argued, factually defensible, visually planned, easy to assemble in DaVinci Resolve, and traceable from idea to published result.
Hermes and generation systems research, draft, rank, suggest, and generate candidates. They must not independently approve their own output. Every decision goes through Mikko.
Every approved idea becomes one episode package containing research, decisions, candidates, approved assets, provenance, handoff info, performance data, and friction notes.
The correct sequence: approved idea → packaging thesis → approved script → claims check → visual beat map → asset requirements → prompts → candidates → selection → recording → Resolve assembly → edit → publish → measure.
Once Mikko approves or locks an item, the system never silently overwrites, mutates, demotes, or replaces it. Regeneration creates a new candidate beside the approved item.
The core production path runs locally. External services are optional alternate lanes, not dependencies. The workflow is supervised — no autonomous publishing.
The pipeline doesn't end at edit. Published performance and production friction feed back into future topic scoring, packaging decisions, and system development.
From idea discovery to published measurement — each stage is approval-gated.
Production coordinator and durable organiser. Collects ideas, generates briefs, maintains episode state, creates visual beat maps, derives asset requirements, generates prompts, records friction.
Local image generation (FLUX.1 Dev) and image-to-video generation (Wan 2.2, 81-frame). Running on RTX 4090 (PRESTO) and RTX 5070 Ti (vidnux). Photorealistic only, no rendered text.
Final editing and creative finishing environment. Mikko is Blackmagic Certified Trainer in Editing, Color, Fairlight, and Fusion. The system prepares and validates media — Resolve is where it comes together.
10GbE NAS serving all Resolve clients. Structured inbox, camera originals, AI-generated media, and staged production assets. ASCII-safe naming enforced for all generated files.
Codex serves as the main implementation agent for code changes. ChatGPT as a drafting and second-opinion layer. All supervised by Hermes and approved by Mikko.
Local-first web UI for episode planning and execution. Package manifests, review views, generation lanes, and friction logging — all in one controlled interface.
Approves: active ideas, topic angle, audience promise, factual position, final script, title, thumbnail, description, selected images, selected videos, presenter take, final edit, publishing, and all durable workflow changes. Operates local hardware, ComfyUI, Resolve, and external generators.
Collects and ranks ideas, generates structured briefs, maintains episode state, prepares candidate sets, preserves decision history, creates visual beat maps, derives asset requirements, generates prompts, prepares Resolve handoffs, records friction, connects performance results back to future planning.
FLUX, Wan 2.2, GPT image systems, Kling, and similar tools produce candidates. They have no approval authority. Every output is tied to episode, script version, beat, originating prompt, generator lane, source asset, status, and approval decision.
The production system is built in the open. Public repositories below.
Local-first YouTube episode planning and execution workflow app. The cockpit UI for the entire production pipeline — package manifests, review views, generation lanes, and friction logging.
Quality assurance tools for creator workflows. Checks and validation for video production pipelines.
A searchable library app for managing and organizing production prompts.
The production system powers every VIDTOOLZ video. Watch the tutorials or explore the code.