The Anti-AI-Slop Design Principles Hallmark Uses to Eliminate Generic AI Output
Hallmark enforces a rigorous anti-AI-slop design philosophy through six universal disciplines—including pre-emit self-critique, locked design tokens, and honest copy—backed by automated slop-test gates that reject generic patterns before they reach production.
Hallmark is an open-source design framework (Nutlope/hallmark) built specifically to combat the "template-ish" outputs common in generative AI. Rather than accepting default aesthetic patterns, the system encodes a strict set of anti-AI-slop design principles that govern every artifact from typography to macrostructure.
The Six Universal Disciplines of Anti-AI-Slop Design
According to SKILL.md (lines 42-56), Hallmark mandates six non-negotiable guardrails that apply to every design flow. These disciplines form the foundation of the framework's quality assurance system.
Pre-Emit Self-Critique
Every artifact must pass a six-axis evaluation before emission: Philosophy, Hierarchy, Execution, Specificity, Restraint, and Variety. As defined in references/slop-test.md, any score below 3 triggers immediate revision. The system stamps each output with a critique signature that documents its ratings:
/* Hallmark · pre‑emit critique: P5 H4 E5 S4 R5 V5 */
Honest Copy and Data Integrity
The framework prohibits fabricated metrics, testimonials, or logos. According to references/anti-patterns.md under the section "Critical (ships as slop) → Invented metrics," all claims must use real numbers or explicit placeholders—never synthetic social proof that characterizes low-quality AI generation.
Locked Design Tokens
Color and font references must resolve to named tokens rather than raw values. The references/anti-patterns.md file explicitly mandates var(--color-accent) and var(--font-display) while rejecting inline OKLCH or hex values, ensuring systematic consistency across the design system.
.button {
background: var(--color-primary);
font-family: var(--font-display);
}
Authentic Visual Assets (No Re-Drawn Chrome)
Hallmark forbids fake browser bars, IDE windows, or phone frames. As noted in references/anti-patterns.md under "Re-drawn chrome," designers must use real screenshots or omit chrome entirely—never re-drawn UI decorations that signal AI generation.
Mobile-First Responsiveness
Every output must render cleanly at 320px, 375px, 414px, and 768px without horizontal scroll, hidden overflow, or broken click targets. The references/responsive.md file mandates strict CSS rules:
/* Non‑negotiables – tested at 320/375/414/768px */
html, body { overflow-x: clip; }
.grid {
display: grid;
grid-template-columns: repeat(auto-fit, minmax(0, 1fr));
}
Typography Purity
Headings must always use Roman typefaces; italics are reserved strictly for body copy. This rule, documented in references/anti-patterns.md under "Italic headers," prevents the common AI tendency to over-style headings with inappropriate oblique type.
h1, h2, h3 {
font-style: normal; /* never italic */
}
Structural Variety and the Macrostructure Catalog
To avoid repetitive "hero → features → CTA → footer" rhythms, references/structure.md defines 21 distinct macrostructures. Each page must deploy a different architectural pattern, ensuring no two outputs follow identical structural progressions.
The Slop-Test Gate and Anti-Pattern Detection
The references/slop-test.md file implements an automated quality gate (checks 46-53) that enforces all core principles before any artifact is emitted. This system acts as the final barrier against anti-AI-slop design violations.
Detecting AI-Generated UI Signatures
references/anti-patterns.md catalogs tell-tale AI aesthetic signatures—including purple-gradient heroes, side-stripe cards, and full-viewport centered heroes—that must be removed or replaced. For example, the canonical "purple-gradient hero" must be replaced with a single-anchor hue:
/*❌ Anti-pattern: purple-gradient hero */
.hero { background: linear-gradient(purple, pink); }
/*✅ Hallmark-compliant: locked token */
.hero {
background: var(--color-neutral-2);
}
Summary
- Hallmark's anti-AI-slop design framework relies on six universal disciplines defined in
SKILL.md(lines 42-56) - Pre-emit self-critique scores artifacts on six axes: Philosophy, Hierarchy, Execution, Specificity, Restraint, and Variety
- Locked tokens (
var(--color-primary)) replace inline hex values; honest copy prohibits fabricated metrics and testimonials - Mobile responsiveness is non-negotiable at 320px, 375px, 414px, and 768px breakpoints with mandatory
overflow-x: clip - The slop-test gate automatically detects and blocks common AI UI patterns like purple gradients, italic headers, and re-drawn chrome
Frequently Asked Questions
What makes a design "AI slop"?
AI slop refers to generic, template-like outputs characterized by repetitive macrostructures, purple-gradient heroes, fake UI chrome, and fabricated metrics. Hallmark identifies these patterns in references/anti-patterns.md and blocks them through automated detection gates that reject artifacts exhibiting these signatures.
How does Hallmark prevent fake data in designs?
The framework enforces Honest Copy as a core principle, requiring real numbers or explicit placeholders while prohibiting invented testimonials, metrics, or logos. Violations fall under "Critical (ships as slop)" anti-patterns and cause immediate failure of the slop-test gate defined in references/slop-test.md.
What are the six scoring axes in Hallmark's self-critique?
Artifacts are scored on Philosophy, Hierarchy, Execution, Specificity, Restraint, and Variety (rated 1-5), with any score below 3 requiring revision before emission. This implementation appears in references/slop-test.md and produces the comment stamp visible in generated code: /* Hallmark · pre‑emit critique: P5 H4 E5 S4 R5 V5 */.
Why does Hallmark require locked tokens instead of hex codes?
Locked tokens ensure design system consistency and prevent the arbitrary color values typical of AI-generated slop. According to references/anti-patterns.md, all references must use named CSS variables like var(--color-accent) rather than inline OKLCH or hex values, maintaining typographic and chromatic discipline across outputs.
Have a question about this repo?
These articles cover the highlights, but your codebase questions are specific. Give your agent direct access to the source. Share this with your agent to get started:
curl -s "https://instagit.com/install.md" Maintain an open-source project? Get it listed too →