The workshop · 12 posts · 92 min
What we learnat the benchwith these models
How the models actually behave, what changed when they did not, and the tests behind every claim. Written by the people building Cetus.
Start here
Four posts in reading order. They assume nothing and they build on each other.
Everything
Twelve posts on prompting, models, cost and craft. Filter by what you are after.
12 posts

One line to sixty minutes: how Director builds a film
Guide · 12 min · 21 Aug 2026

Start frames beat adjectives
Craft · 6 min · 14 Aug 2026

Choosing a model: speed, cost and what each one is good at
Comparison · 11 min · 7 Aug 2026

A hundred videos in one run without babysitting it
Guide · 8 min · 31 Jul 2026

Credits, explained: what each render costs and why
Release · 5 min · 24 Jul 2026

Keeping a character the same person across twelve shots
Craft · 7 min · 17 Jul 2026

Writing a shot list the models can actually render
Guide · 9 min · 10 Jul 2026

The eight-second problem, and how to cut around it
Craft · 6 min · 3 Jul 2026

Voice is free. Here is how to use it like it matters.
Guide · 6 min · 26 Jun 2026

What we changed after watching two hundred failed renders
Release · 8 min · 19 Jun 2026

Framing for a destination, not for a canvas
Craft · 5 min · 12 Jun 2026
How a claim gets in here
Four steps, and the fourth is the one most technique writing skips.
Step 1
One prompt, one variable
The scene is written once. Every later version changes exactly one clause, so a difference on screen has one possible cause.
Step 2
Five seeds a version
One render tells you what a model did once. Five tell you what it does. Anything that only happens once is not a finding.
Step 3
Watched, not skimmed
Every clip is watched to the end at full size. A quarter of what fails does so in the last two seconds, where a poster frame cannot see it.
Step 4
Written down either way
The things that changed nothing are in the posts too. A technique list with no negative results is somebody's habits, not a test.
Reading is the slow way
Every prompt in here is one you can run
The prompt library carries the sentence behind each published result, with the model it needs already chosen. Free to read and copy, signed in or not.
Browse prompts
