Build a Show LUT for Mixed AI Footage
A CDL-plus-LUT recipe so mixed-model shots sit on one timeline, a stills chart to test it, and the rule for when to regenerate instead.
A CDL-plus-LUT recipe so mixed-model shots sit on one timeline, a stills chart to test it, and the rule for when to regenerate instead.
Keep DX, MX, FX and optional ambience as separate files, with a sync pop, so a later-language dub is a voice swap, not a remix.
When a social cut has no silence for standard AD: freeze, extend, or ship a second described cut, and brief the editor so both versions stay in sync.
Review the localized cut on picture with visible time, using a scorecard that splits language, culture, and legal. A script doc misses collisions.
Give mixers timecode, stem, direction, amount in dB or Hz, and a reference clip. A one-page template that replaces 'make it cinematic'.
Most briefs lose their constraints at the first handoff. A template that carries locks, references and rejected directions, and stays next to the review loop.
A look LUT on unnormalised clips keeps the mismatch and tints it. Neutralise each shot, match to a master, then grade for style.
Tests without a written record get re-run every quarter at full cost. Log these fields per test, then review on a rhythm that feeds a hypothesis backlog.
A 20-minute agenda that turns last week's curves into one edit decision, one packaging decision, and three numbers you log so the trend is visible.
Long renders don't have to freeze the conversation. How to hand a slow multi-step job to a background task and keep briefing the agent on the next scene.
Clips per source hour is the repurposing metric most teams never measure. How to calculate current yield, a realistic target, and the steps that move it.
A format is a fixed structure plus one variable. Eight of them, what each demands in source material and production weight, and which workflow to build it on.
A flat list of 139 tools serves an agent worse than six domain specialists. Name the job so the router sends the turn to the right one first.
A background task name is the saved-workflow key. Name it by intent so a later re-run updates one recipe instead of spawning a v2 row.
Price by the build hours a pack removes, not file count. A positioning test for your own pack, plus a bundle structure that lifts order value without discounts.
Run provider-check for every model in a batch before the first POST, so an unroutable name fails the preflight instead of item 87.
No camera, no studio, no on-camera talent, and in some cases no product yet. A launch video plan that names the substitute for every missing input.
Each of the eight installed skills covers a different pipeline shape. Leave knowing which one an agent will reach for, and how to phrase the request.
Schedules match five-field cron against UTC. A local posting time pinned once will drift twice a year unless you rewrite the expression.
Generation is fast. Client approval isn't. A review loop built from free previews, no-login share links, and a QC pass before anything gets sent.
The instruction that survives forty different photos, the cheap model that makes a bad first pass affordable, and how much of the batch to actually check.
A client's small tweak and a client's swap-the-look request aren't the same change order. A credit taxonomy for what each revision type actually costs.
Every talking-head video already contains its own transcript. What it's actually raw material for, and why the text file outlasts the burned-in caption.