Pipelines
I don't automate, I collaborate. My pipelines aren't programs, they're conversations - I work with AI agents to build them, not just run them. Creativity and efficiency both get rewarded, and AI slop gets called out and cut the moment it shows up.
The Pieces of My Pipeline
Four phases, borrowed from film
Development
Story/Audience/Deliverables/Formats
What we are making, who it is for, what formats the deliverables need. All of it decided before a single token is spent - the one-page brief everything downstream points back to.
Pre-production
Model selection/Storyboards/Reference sheets/LoRA training/Music
Which models serve the project, and what they cost, shared openly. Storyboards iterated fast and locked. Custom models trained and approved. Music often gets made here too. Where token management pays off most.
Production
Generation/Dailies/Approve / re-shoot
Footage generated in ComfyUI from the locked references, master soundtracks built. Reviewed daily, approved or re-shot. Time and budget gate every generation, so the production never goes off the rails.
Post-production
ComfyUI → Resolve bridge/Edit/VFX / titles / sound/Grade/Render
Edit, VFX, titles, sound, atmosphere, colour grade, render. Rough cut, pre-final, final.