About

Staring at a blank canvas without direction leads to decision paralysis. This generator constructs syntactically valid English sentences using a generative grammar engine built on phrase-structure rules. The lexicon is curated from frequency-ranked corpora including the Google Web Trillion Word Corpus and the American English Subtitle Corpus (SUBTLEX-US), filtered to retain words with strong visual or conceptual associations. Each prompt is assembled recursively: a random sentence template is selected based on a complexity parameter C ∈ {1, 2, 3}, then noun phrases, verb phrases, adjective phrases, and prepositional phrases are resolved by sampling from subcategory pools. The result is not a keyword list. It is a grammatically coherent scene description that forces compositional decisions.

Complexity level 1 yields simple subject-verb-object structures with common vocabulary. Level 3 introduces nested prepositional phrases, compound modifiers, and rarer vocabulary drawn from the long tail of the frequency distribution. The tool applies English morphology rules for pluralization and article selection (a vs. an) at generation time. Note: prompts are syntactically valid but semantically surreal by design. That is the point. Constraints breed creativity.

Formulas

The generator follows a recursive phrase-structure grammar. A sentence S is expanded from a template T selected at random from the set of templates matching complexity C:

S ← resolve(T_C)

Each template is a sequence of tokens. A token is either a literal string or a phrase reference {category}. Resolution is recursive:

resolve(token) =

{

token if literalpick(category, C, args) if phrase ref

The pick function selects a random entry from the lexicon pool filtered by category and args (subcategory), weighted by complexity level. Article insertion uses phoneme analysis:

article(w) =

{

"an" if isVowelSound(w[0])"a" otherwise

Pluralization applies English morphological rules:

plural(w) =

{

w + "es" if ends in s, x, z, ch, shw[0..n−1] + "ies" if ends in consonant + yirregular[w] if in irregular mapw + "s" otherwise

Where w is the input word, n is its length, and irregular is a lookup table of common irregular plurals (child → children, mouse → mice, etc.).

Reference Data

Phrase Type	Role in Prompt	Example Words (Level 1)	Example Words (Level 3)	Subcategories
Noun Phrase	Subject / Object	cat, mountain, river	automaton, labyrinth, chrysalis	animal, place, object, abstract, person
Adjective Phrase	Modifier	red, old, bright	iridescent, decrepit, luminous	color, size, texture, mood, age
Verb Phrase	Action	runs, holds, watches	dissolves, fractures, entangles	motion, creation, destruction, perception
Prepositional Phrase	Scene context	in the forest, on a hill	beneath a fractured dome, amid swirling ash	location, temporal, manner
Adverb Phrase	Action modifier	slowly, quietly	reluctantly, imperceptibly	speed, manner, degree
Complexity Level Reference
Level 1	Simple SVO. 5 - 10 words. Common vocabulary. Single clause.
Level 2	Compound sentences. 10 - 18 words. Mixed-frequency vocabulary. One prepositional phrase.
Level 3	Complex nested structure. 15 - 30 words. Rare vocabulary. Multiple modifiers and clauses.
Lexicon Statistics
Total Nouns	200+ across 5 subcategories
Total Adjectives	150+ across 5 subcategories
Total Verbs	100+ across 4 subcategories
Preposition Templates	60+ across 3 subcategories
Sentence Templates	18 templates across 3 complexity tiers

Frequently Asked Questions

Complexity C controls two axes: sentence structure and vocabulary rarity. Level 1 uses simple subject-verb-object templates with high-frequency words (top 5000 in corpus ranking). Level 2 introduces compound clauses and prepositional phrases, drawing from the top 15000. Level 3 nests multiple phrases and uses rare, evocative words beyond rank 15000. Higher complexity yields longer, more surreal prompts that challenge compositional skill.

Syntactically, yes. The phrase-structure grammar enforces subject-verb agreement, correct article usage (a vs. an), and proper pluralization. Semantically, the prompts are intentionally surreal. A sentence like "a luminous fox dissolves beneath a fractured dome" is grammatically valid but describes an impossible scene. This semantic tension is the creative catalyst.

Extremely unlikely. With 200+ nouns, 150+ adjectives, 100+ verbs, and 18 templates, the combinatorial space exceeds 10⁸ unique outputs at complexity level 2 alone. The generator also tracks the last 100 prompts in history and avoids exact repeats within that window.

The lexicon draws from the Google Web Trillion Word Corpus for written-English frequency ranks and SUBTLEX-US (American English subtitle corpus) for spoken-English naturalness. Words were filtered to retain only those with strong visual or sensory associations. Abstract or purely technical terms (e.g., "amortization") were excluded in favor of concrete, paintable concepts (e.g., "chrysalis", "ravine").

The generator checks against a curated exception list. Words starting with a silent h ("hour", "honest") receive "an". Words starting with a pronounced u producing a /juː/ sound ("uniform", "unicorn") receive "a" despite the vowel letter. The exception list covers approximately 40 common edge cases drawn from English phonology.

Yes. When generating a batch of N prompts, the generator maintains a per-batch exclusion set. Each resolved sentence is checked against all previously generated sentences in the same batch. If a collision occurs (identical string), the generator re-rolls up to 5 times before accepting a near-duplicate. Additionally, the template selector cycles through available templates before repeating one within a batch.