About

Standard Lorem Ipsum derives from Cicero's De Finibus Bonorum et Malorum (45 BC) and carries zero semantic signal for modern layout work. Cat Ipsum replaces it with behaviorally accurate feline prose. Each sentence is assembled from a grammar engine that combines 14 syntactic templates with categorized lexicons of cat behaviors, vocalizations, and environmental interactions. The output reads as coherent (if absurd) narrative rather than shuffled word salad. This matters because placeholder text that resembles real content exposes typography and layout problems that random Latin obscures.

The generator uses weighted random selection with a recency buffer of depth 3 to prevent immediate phrase repetition. Paragraph length follows a bounded pseudo-Gaussian distribution centered on 5 sentences with σ = 1.5, clamped to [3, 8]. Three style modes adjust vocabulary density: "Lazy" favors sleeping and lounging verbs, "Chaotic" biases toward destruction and zoomies, and "Refined" selects grooming and judgmental observation patterns. Output is pure text with no markup injection, safe for direct paste into any CMS or design tool.

Formulas

Each sentence is constructed from a template T_i selected from the template pool of size n. Slot variables within each template are filled by weighted random sampling from categorized lexicons L_k.

S = fill(T_i, L₁, L₂, …, L_k)

Where template index i is drawn from a uniform distribution excluding the last 3 used indices (recency buffer B):

i ∼ U(0, n) ∖ B

Paragraph length p follows a clamped Gaussian:

p = clamp(round(μ + σ ⋅ Z), 3, 8)

where μ = 5, σ = 1.5, and Z is a standard normal variate generated via the Box-Muller transform:

Z = √−2 ln U₁ ⋅ cos(2πU₂)

where U₁, U₂ ∼ U(0, 1).

Where S = generated sentence, T_i = template at index i, L_k = lexicon category k, B = recency buffer (last 3 indices), p = sentences per paragraph, μ = mean paragraph length, σ = standard deviation, Z = standard normal variate.

Reference Data

Cat Behavior	Frequency	Typical Duration	Destructive Potential	Human Annoyance Level
Sleeping on keyboard	High	2-4 hours	Low (data loss: Medium)	Moderate
Knocking objects off table	High	5-30 seconds per item	High	High
3 AM zoomies	Daily	10-45 minutes	Medium	Extreme
Ignoring expensive toy	Constant	Permanent	None	Moderate
Playing with cardboard box	High	1-3 hours	Low	None
Sitting in doorway undecided	Moderate	1-10 minutes	None	High
Grooming aggressively	Moderate	15-45 minutes	None (hairball: Medium)	Low
Staring at wall	Moderate	5-30 minutes	None (paranoia: High)	Unsettling
Bringing dead prey indoors	Low - Moderate	Event-based	High (biohazard)	Extreme
Demand feeding at 4 AM	Daily	Until fed	Low	Extreme
Sitting on laptop	High	10-60 minutes	Medium (overheating)	High
Chattering at birds	Moderate	2-10 minutes	None	Amusing
Refusing to move from chair	High	Hours	None	Moderate
Sudden belly trap	Moderate	3-5 seconds	Medium (scratches)	Painful
Hiding in paper bag	Moderate	15-60 minutes	Low	None
Kneading blankets	High	5-20 minutes	Low (fabric pulls)	Endearing
Slow blinking	Moderate	Momentary	None	Heartwarming
Sprinting sideways	Low	5-15 seconds	Low	Hilarious
Eating houseplants	Low - Moderate	1-5 minutes	High (toxicity risk)	High
Loafing	High	30-120 minutes	None	None

Frequently Asked Questions

Cat Ipsum uses template-based grammar construction rather than naive word shuffling. Each sentence follows a valid subject-verb-object pattern drawn from 14 syntactic templates, with slots filled from categorized lexicons (behaviors, sounds, body parts, objects). The result reads as coherent prose with comedic feline logic, not gibberish. A recency buffer of depth 3 prevents immediate template or phrase repetition, ensuring variety across paragraphs.

Paragraph length follows a clamped pseudo-Gaussian distribution using the Box-Muller transform. The mean is 5 sentences with a standard deviation of 1.5, clamped to the range [3, 8]. This produces natural-feeling length variation that mimics real editorial text rather than rigid fixed-length blocks.

Each style mode applies a different weight map to the lexicon categories. "Lazy" mode increases selection probability for sleeping, lounging, sunbeam, and ignoring-related vocabulary by a factor of 3×. "Chaotic" mode boosts zoomies, destruction, knocking, and 3 AM behavior terms. "Refined" mode favors grooming, judging, slow-blinking, and sophisticated disdain vocabulary. The templates themselves remain the same across modes - only the lexicon sampling weights change.

Yes. The output is pure plaintext with no HTML tags, script injections, or special characters beyond standard punctuation. It is safe for direct paste into WordPress, Figma, Sketch, Adobe InDesign, or any CMS text field. Paragraph breaks use standard newline characters.

Mathematically, yes - the combinatorial space is large but finite. With 14 templates and lexicons averaging 20-40 entries per category, the theoretical sentence space exceeds 10 million unique combinations. However, within a single generation, the recency buffer actively prevents the same template from appearing within 3 consecutive sentences. Across separate generations, the PRNG seed differs, making exact repetition statistically unlikely but not impossible.

A true Markov chain requires a large training corpus and produces output quality that degrades unpredictably - sometimes generating nonsensical fragments or plagiarized passages. Template-based generation guarantees grammatical correctness on every output while maintaining full control over humor and tone. The trade-off is slightly less organic variation, offset by a much larger and more reliable lexicon than a small-corpus Markov model would produce.