My AI's first try at my editing style was sloppy. That was my word for it. So it got a list of things it's never allowed to put on screen, and every frame gets checked against it.
Read https://jay-burnett-packs.vercel.app/killlist/agent.md and set this up for me.It reads agent.md: install, test and undo steps. Want it by hand instead? Your email below opens the full page.
You're in, the files are below.
Everything my AI editor is banned from putting on screen, with a one-line reason each.
# The kill list: what my AI editor is never allowed to put on screen
Each line has the reason. If a frame breaks one, the frame gets rebuilt before I see it.
## Motion
- **Graphics that don't match the sentence being said.** Mute test: turn the voice off and the graphic should still say something. Motion with no message is the number one AI tell.
- **One easing curve on everything.** Real editors use several (fast in, soft out, different for moving UI). Same curve everywhere reads as a template.
- **Bounce and overshoot on every element.** Only the one hero beat in a scene gets it, and a counter never overshoots its number.
- **Everything moving at once.** One motion source at a time, so the eye knows where to go.
- **Crossfades between talking-head cuts.** Hard cuts on the beat. Crossfades are only for time passing.
## Layout and type
- **Big centered all-caps headlines as the whole style.** It's the default AI look. One title per section, then small captions.
- **Kinetic caps slams, scoreboard type ("PASS 8.8/10"), HUD corner labels, floating checklist widgets.** They fill space and say nothing.
- **Screenshots tilted, with a rounded card outline, a white border or a drop-shadow "polaroid".** Drop real screens in flat: no tilt, no outline, no border, no shadow.
- **Sticky notes and fake UI.** No placeholder screens. Real screens with real, non-round numbers.
- **Bento grids, starbursts, sparkles, default icon packs.** Instant template look.
- **Purple and blue gradients, gradient text, glow and neon.** The most common AI palette there is.
- **Huge captions with punctuation.** Captions stay small, 1 to 3 words, no punctuation.
## Proof
- **No real asset for more than about 5 seconds.** Something real (my screen, my file, my number) shows up at least every 5 seconds.
- **Mixed looks in one video.** One look per video. Light editorial, dark and boxed captions never share a video.
Every frame answers these before I see it. Any no means the frame gets rebuilt.
# The frame check: every frame passes this before I see it. Any "no" means rebuild the frame.
1. Does the graphic answer the exact sentence being spoken right then?
2. Is there ONE look across the whole video?
3. Are screenshots and images dropped in flat (no tilt, no card outline, no white border, no polaroid shadow)?
4. Is it free of kinetic caps slams, scoreboard type, HUD corner labels and floating checklists?
5. Are captions small, 1 to 3 words, outlined, no punctuation, never italic serif?
6. Is the layout the one style I picked, with one graphic block and one title at most?
7. Is there a real asset (my screen, my timeline, a real number) at least every 5 seconds?
8. Has a second AI that didn't build it scored the frame, with its verdict shown?
9. Has a reviewer that knows my channel checked it, and was every note fixed or answered?
Two boards from my own renders, not the same frame redone. Left: the first style board, the one I called sloppy (big all-caps headline, a terminal line floating on my face). Right: a later board I approved, sentence case with one accent word and a real photo dropped in flat, no tilt and no border.

