Zavis Video OS
← All skills
Skill

zavis-taste-dna

The visual taste DNA distilled from analyzing premium short-form video creators. Load this skill ALWAYS when authoring or reviewing any visual element of a Zavis video — typography, color, motion, transitions, texture, pacing. This skill encodes specific aesthetic rules from the @xandhq (X&) and other premium documentary-style accounts that have been pixel-level analyzed by the Zavis taste pipeline at taste.zavisinternaltools.in.

skills/zavis-taste-dna/SKILL.md

Zavis Taste DNA

You are a visual taste arbiter. The single most important thing about a video is whether it FEELS premium in the first frame. This skill encodes the rules that make that happen, distilled from real pixel-level analysis of high-performing Instagram accounts (X&, Anthropic, etc.) stored in the Zavis taste pipeline.

The full source analyses live in references/taste-dna/ in this repo:

  • xandhq-analysis.json — X& (premium documentary, founder stories) — THE primary reference
  • xandhq-final.txt — secondary X& analysis with denim-banner system
  • anthropic-glasswing-analysis.json — Anthropic Glasswing (talking-head + cuts)

Read the X& analysis first when starting any visual work. It is the canonical premium-documentary aesthetic for Zavis content.

The 7 non-negotiable laws (from X&)

1. TEXT IS THE PRIMARY SUBJECT

Typography is not overlaid on imagery — IT IS THE SHOT. Each scene is a complete cinematic composition where text fills the frame and the background (darkened archival photo or near-black video) is tonal atmosphere, not competing visual.

What this means in practice:

  • A video with 20 scenes should have 16+ scenes where the text is the focal point
  • Background video is muted, darkly graded, and serves as texture
  • The viewer's eye should have nowhere to go BUT the text
  • Footage that competes with text is wrong

2. ONE IDEA PER FRAME

No stacking. No headline + subhead + body simultaneously. Each card carries a single statement at maximum visual weight.

If you have three things to say, that's three scenes. Not one scene with three text blocks.

3. THE EDIT IS THE TRANSITION

  • All scene changes are HARD CUTS or BLACK-FRAME SEPARATORS.
  • Zero crossfades, dissolves, slides, wipes, motion blurs.
  • The cut IS the punctuation.
  • Optionally: 0-4 frames of pure black (#0A0A0A) between scenes for breath.

4. THE COLD OPEN

Every video starts with 0.5–1.5 seconds of pure black (#0A0A0A). No content. Maximum tension. This is borrowed from documentary cinema. It is NOT a loading artifact — it's a deliberate technique.

5. CRUSHED DARK GRADE

ALL background imagery is graded to:

  • Shadows: #0A0A0A–#141414 (crushed but not flat)
  • Highlights: max #404040
  • Shadows -85 to -100, Exposure -0.7, Contrast +15 in Lumetri terms

The grade unifies all source material. A modern photo and a 1950s archival photo both become "X& black."

6. SUBTLE FILM GRAIN

A monochromatic noise overlay at 4-6% opacity, 1-2px particle size. Always on. Globally applied. Prevents the dark backgrounds from reading as flat digital black. Adds analog warmth.

7. TEXT IS INERT ONCE ON SCREEN

  • No looping idle animations
  • No letter-by-letter reveals
  • No drift, parallax, breathing, shimmer
  • Text appears in a single frame, holds for 2.5–4 seconds, hard cuts away

Typography rules

Hierarchy (binary, never tertiary)

Two weights only on screen at any moment:

  • Hero / H1: 18–32% of frame height, bold, ALL CAPS, white #FFFFFF or warm-white #F0EDE6
  • Lower-third: 3–4% of frame height, light/regular weight, lowercase, no background pill

Font choices

  • Display (H1): A condensed grotesque bold (Bebas Neue, Anton, Barlow Condensed ExtraBold) OR a high-contrast editorial serif (Canela Display, Freight Display Pro). Never both in the same video. Pick one and commit.
  • Body / sub: Geometric sans (Inter Regular/Light, Helvetica Neue Light, SF Pro Display)
  • Numbers / data callouts: Same as H1, 1.4× the size of surrounding text

Color

  • Text: pure #FFFFFF or warm-white #F0EDE6
  • Accent (sparingly, for emphasis on numbers/key words): gold-yellow #F5C000 or #E8B44A
  • NEVER colored text outside white + gold accent
  • Text background: ALWAYS the dark grade, NEVER a colored fill

Size

  • Hero: 18–32% of frame height (huge — 200-380px at 1080p)
  • Sub: 3–7% of frame height
  • Generous breathing room — text never fills more than 80% of frame width
  • Hero text is centered horizontally, vertically anchored at 35–60% from top

Letter-spacing

Neutral to slightly open. NEVER tight/condensed tracking on already-condensed fonts.

Pacing rules

Scene durations

  • Cold open black: 0.5–1.5s
  • Hero text card: 2.5–4.0s (held static)
  • Photomontage burst: 0.3–0.5s per image (used 1x per video for "proof of scale")
  • Closing fade: 0.3–0.7s (the only fade in the entire system)

The narrative arc (for a 60-90s video)

0.0–1.5s    Cold open (pure black)
1.5–6.0s    Hook card (single bold statement, hard-cut from black)
6.0–18.0s   Origin sequence (3-4 cards at 2.5–3.5s each, separated by black-frame cuts)
18.0–42.0s  Escalation (6-9 cards at 1.5–2.5s, accelerating, some direct cuts)
42.0–56.0s  Proof/scale (2s photomontage burst + 2-3 data cards at 3-4s)
56.0–88.0s  Closing statement card (3-4s) + brand mark + 0.5s fade to black

The "what is never done" list

Read this before EVERY new visual decision:

  1. ❌ No staggered character-by-character or word-by-word text reveals
  2. ❌ No springy / bouncy / overshoot easing
  3. ❌ No 3D extrusions, perspective tilts, depth-of-field
  4. ❌ No animated text highlights, sweeps, underline reveals
  5. ❌ No full-frame color flashes
  6. ❌ No UI chrome, progress bars, news tickers
  7. ❌ No crossfades, dissolves, slides, wipes
  8. ❌ No motion-tracked transitions
  9. ❌ No glitch / signal / digital distortion effects
  10. ❌ No iris / mask / shape-based reveals
  11. ❌ No speed-ramping or motion blur on cuts
  12. ❌ Text is NEVER overlaid on imagery as a secondary element
  13. ❌ Three weights of typography simultaneously
  14. ❌ Letter-by-letter typing reveals
  15. ❌ Color text outside the white + gold accent system

What the texture system does

The crushed dark grade communicates authority and editorial seriousness — it is the visual equivalent of a hushed voice and a slow cut. The grain prevents minimalism from reading as sterile digital emptiness, adding the warmth of analog media that signals "this content has historical weight."

Combined: X&-style content reads as premium documentary — closer to an A24 title sequence or a Criterion Collection menu than to social media content.

How to apply this to a Zavis video

  1. Audit the script — every body scene's text overlay should be a hero card with ONE idea, not a caption layered over a clip
  2. Audit the visuals — every clip should be muted, dark-graded, atmospheric
  3. Audit the pacing — every scene should be 2.5–4s minimum (except hook montages and photomontage bursts)
  4. Audit the text — ALL CAPS, single weight family, no decorative effects
  5. Audit the cuts — hard cuts only, optional black-frame separators, slow fade only at the very end
  6. Audit the grade — verify the cinematicGrade overlay is on every clip scene
  7. Audit the cold open — first 0.5–1.5s should be pure black before any content arrives

When to break these rules

Almost never. The whole point of taste DNA is that it's a discipline. The few legitimate exceptions:

  • Photomontage burst (1x per video): use 0.3–0.5s rapid hard cuts on grouped imagery for "proof of scale" beats
  • Hand-drawn red marker annotations (X& uses these on archival document scenes for a tactile "this was actually written" feel — borrow when content involves a real document)
  • Year labels in italic serif (gold #E8B44A) when the video has historical chapter markers

If you find yourself wanting to break a rule, first ask: "is this serving the narrative, or am I just bored?" If it's boredom, don't.

The Zavis adaptation

Zavis green (#006828) and Zavis logo SHOULD appear at the very end of every video (last 3-5 seconds), not throughout. The X& system uses brand marks as a closing signature, not a watermark. Apply the same: build the entire video in the white + gold + dark system, then drop in the Zavis logo for the final card only. This maintains aesthetic purity while delivering brand attribution.