CVRL0SGSD MediaSubscribe

The method

Anti-slop

The list of things that make a video look generated, and the checks that keep them out. Read before proposing a direction and again before reporting a build.

Not on this channel

  • Hyperactive kinetic captions. Word-by-word pops, bounces, color cycling, emoji inserts. The two-word uppercase captions are the caption treatment; they do not animate.
  • Decorative 3D. Floating product renders, spinning devices, glass spheres, isometric desks. The real product on the real desk is on camera.
  • Arbitrary gradients and glows. No gradient text, no radial glow behind a card, no neon edge, no lens flare. Grounds are footage, flat ink, or flat paper.
  • Unmotivated motion. Breathing backgrounds, drifting ghost text, parallax layers, pulsing accents, orbit rings, particles. If it moves while nothing is being said about it, it is out.
  • Fake UI and fake data. Mocked notifications, invented charts, placeholder numbers, rounded figures presented as measurements, dashboards nobody opened. A number on screen was measured or paid, and the narration says so.
  • Generated stand-ins for real evidence. An AI image of “a worn wallet” when the worn wallet is on the desk. A rendered “unboxing” when the box exists.
  • Attention hacks that distort the point of view. Countdown timers, “wait for it”, red arrows and circles, zoom-punches on a reaction, a score the narration never gave.
  • Creator clichés. Subscribe-button animations, bell icons, “like if you agree”, pointing at a floating caption, split-screen with a stock reaction.
  • Stock texture as brand. Film grain overlays, light leaks, VHS lines, paper textures on cards. The desk supplies the texture.
  • Banned faces. Inter and Space Grotesk as visible type. Both are among the most-flagged default faces of 2025–2026. Fine as invisible system fallbacks, never on screen.
  • Second accent colors. Gold is the accent. A “warning red” for the catch is still a second accent.

A vendored default this repo overrides

The vendored house style of one graphics toolkit tells an agent that “every scene needs visual depth”, meaning two to five ambient decoratives: glows, ghost text, grain, slow breathing motion. It calls one decorative “under-dressed”.

That guidance does not apply here. On this channel a scene is the footage; depth comes from the desk. Where the vendored skill defers to a project design spec, the project spec wins.

Self-check before a direction or a report

Answer each in one line. A “no” on any of the first four stops the build.

  1. Does real footage or a real photo exist for this beat? If yes, is it the foundation of the shot?
  2. Does the graphic do one of the four jobs (clarify, context, relationship, pause)? Which?
  3. Does the graphic claim exactly what the narration claims, with the same hedge, and no more?
  4. Is every motion cited from the four moves, or written up as an exception with a reason?
  5. Count the things on screen. Could one come out without losing the point?
  6. Is gold used once, on the thing that matters?
  7. Would this frame look at home in a product launch? If yes, strip it back until it looks like a notebook page.

When in doubt

Static shot, hard cut, one label. Then ask.

Part of the method, v1.0. Revisions are recorded in what changed.