Brand Voice

The single biggest lever on generation quality is the brand-voice config. Find it in Project settings → General.

Tone

Pick 3–5 adjectives. Examples:

• playful, irreverent, witty
• expert, authoritative, calm
• aspirational, premium, confident
• educational, patient, encouraging

The model uses these as anchors. Specific is better than generic - "playful" alone is ambiguous; "playful, irreverent, slightly chaotic" is clearer.

Banned phrases

Hard deny list. Generated content containing any banned phrase is auto-rejected and regenerated. Common bans:

• Compliance triggers - "guaranteed", "best in class", "FDA-approved" for unregulated products, "100%".
• Legal triggers - competitor brand names you can't mention.
• Brand-tone violations - clichés you hate ("level up", "game-changer").
• Platform-policy traps - "click here" (Instagram), "FREE!!!" (everywhere).

Reference posts

Paste URLs of:

• Past winners of yours - what hit hard, what to repeat.
• Competitors doing it well - what to study (the model studies pacing, hook structure, length).
• Tone exemplars from outside your category - a brand whose voice you admire.

Brand-specific terms

Add your product names, taglines, jargon. Layers respects exact casing ("Acme" not "ACME" or "acme") and keeps these intact across regenerations.

Visual identity

In media settings:

• Upload your logo (transparent PNG or SVG).
• Upload color reference imagery - the model samples palettes for image generation.
• Tag a "hero" image set for product-focused content.

Audience

In project settings:

• Demographics - age range, gender mix, geos.
• Interests - top 5 things this audience cares about.
• Pain points - problems they want solved.
• Preferred tone - formal vs casual.

Iterating

Edit voice config and observe the next batch of generations. The new config applies to subsequent generations only - content already generated isn't regenerated automatically. Regenerate individual drafts from their detail panel if you want them on the new voice.