Brand Color and Typography Locks for AI Explainer Series
Freeze palette and type rules that survive generative models so multi-week AI explainer series stay on-brand—plus quick QA swatches.

Brand Color and Typography Locks for AI Explainer Series
Keep AI-generated teach video on-brand across a multi-week series by freezing palette and type rules that survive generative drift—not by hoping each episode “feels close enough.”
Brand and marketing teams shipping stylized or host-led explainers hit the same failure: episode one matches the deck; episode four invents a cousin palette and a heavier title face. Style locks for weekly batches help (one approved style lock for a week of explainers); this article focuses on the color and typography contracts inside that lock.
What to freeze vs allow per episode
| Layer | Freeze across series | Allow per episode |
|---|---|---|
| Primary / secondary / accent hex | Yes | Seasonal accent only if pre-approved |
| Background family (light/dark) | Yes | Scene metaphor content |
| Title face + weight | Yes | Title words |
| Body / caption face | Yes | Caption length |
| Corner radius / card chrome | Yes | Card content |
| Host wardrobe palette | Yes (range) | Pose / gesture |
| Illustration metaphor | Direction only | Specific objects |
| Motion language | Easing / wipe set | Which beats move |
If a model regenerates “teal-ish” instead of your hex, treat it as a fail—not a vibe.
Palette rules that survive gen models
Generative image/video models approximate color. Your lock must be enforceable in QA, not only in prompts.
Practical rules:
- Name three hexes — primary, secondary, accent — plus forbidden near-misses (“no purple-teal,” “no neon coral”)
- Specify background luminance — light series vs dark series; do not mix mid-series
- Limit gradients — flat or one approved gradient; random mesh gradients break brand systems
- Caption contrast — require WCAG-minded contrast for burned-in text; see accessibility checklist for educational Reels
- UI skin — if you show product UI, decide: true product colors vs branded stylized UI—and stick to one
- Error / success colors — lock semantic colors so “danger” never appears as brand accent by accident
Prompt text alone is weak. Attach a style board still with swatches visible in-frame for regenerations, or post-process with a controlled grade. Character wardrobe should reference the same triad (reusable AI character bible).
Quick QA swatches
Before approving an episode:
- Drop primary/secondary/accent chips next to three sampled frames
- Reject if accent appears as a different hue family
- Reject if backgrounds shift from light series to charcoal mid-arc
- Reject if cards gain unapproved glow or glass effects
- Reject if success/error semantics collide with brand accent
One reviewer with a printed or on-screen swatch strip beats five Slack opinions. Keep a reject folder of near-miss frames; it trains vendors faster than adjectives.
Typography locks for vertical teach
Type fails more quietly than color. Models invent faces; editors “fix” with random bold.
Lock:
- Title: one family, one or two weights, max size band for 9:16
- Caption / step text: one family optimized for small size and mute-first reading
- Numerals: tabular if you show steps or metrics
- All-caps policy: never / titles only / never in body
- Tracking and line length: short lines; avoid paragraph blocks on screen
- Language variants: if you ship bilingual captions, confirm glyph coverage for both scripts before locking a face
Pair with safe zones and crop planning for 9:16 so type never sits under platform UI.
When stylized looks fight trust—especially in regulated niches—prefer clearer type over decorative faces (when stylized looks hurt trust). Whiteboard vs flat vector vs clay choices also interact with type weight (style by concept type).
Caption hierarchy example
- Hook line: title face, largest approved size
- Step labels: caption face, medium, numbered
- Fine print / disclaimer: caption face, smallest approved size, highest contrast
Do not invent a fourth face for “emphasis.” Use weight or color from the locked set.
Series governance
- Version the style lock (
brand-explainer-v3) and date it - Prohibit mid-series “refresh” without a new version bump
- Keep a reject reel of near-miss frames for vendors and models
- Align host locks with the character bible so wardrobe colors do not fight the palette
- Tie seasonal layer swaps to the same lock (seasonal campaigns without full rerenders)
Vendor and model brief insert
Paste into every generative brief:
- Hex triad + forbidden hues
- Title/caption face names and specimen link
- “Reject near-miss color or substitute fonts”
- Link to style board still
If vendors deliver drift, the reject folder is your contract language. Weekly style-lock batches stay cheaper than episodic debates.
Soft production tip
Lock color and type with the style board so regenerations do not drift. In storyboard-first workflows (Scro included), the approved board carries swatches and type specimens beside beats—cheaper to enforce than grading ten finished clips.
One-page lock template
- Series name / version / owner
- Hex triad + forbidden hues + semantic colors
- BG mode (light/dark)
- Title face + caption face + samples at 9:16
- Card chrome rules
- QA swatch ritual (who signs)
- Link to character bible + claims rules if any
Brand consistency in AI explainers is an ops problem: freeze what must not move, allow only the teach content to change, and reject near-misses early.
