Index/writing
House · writers, poets, screenwriters, narrative designers
✍️

Writing, Poetry & Narrative

100
Entries
4
Primitives

Showing 100 of 100 entries

seasonal haiku craft№ a-0

Haiku Breath LoRA

Fine-tune Qwen/Qwen3.5-4B on 90 (season, haiku) pairs so the model breathes kigo-accurate micro-poems.

SFT · Instruction→
hardboiled narration№ a-1

Noir Voice LoRA

Fine-tune Qwen/Qwen3-8B on 80 (scene, narration) pairs so the model writes chandlerian detectives.

SFT · Instruction→
screenplay structure№ a-2

Beat Sheet LoRA

Fine-tune Qwen/Qwen3-8B on 70 (logline, beats) pairs so the model drafts three-act outlines.

SFT · Instruction→
classical simile craft№ a-3

Epic Simile LoRA

Fine-tune Qwen/Qwen3.5-4B on 60 (action, simile) pairs so the model extends homeric comparisons.

SFT · Instruction→
narrative game prompts№ a-4

TTRPG Pact LoRA

Fine-tune Qwen/Qwen3-8B on 100 (hook, module) pairs so the model writes tabletop one-shots.

SFT · Instruction→
iambic pentameter№ a-5

Sonnnet Meter LoRA

Fine-tune Qwen/Qwen3.5-4B on 85 (thought, sonnet) pairs so the model scans strict十四行.

SFT · Instruction→
tv cold opens№ a-6

Cold Open LoRA

Fine-tune Qwen/Qwen3-8B on 75 (premise, open) pairs so the model scripts punchy teases.

SFT · Instruction→
flash fiction№ a-7

Flash Foam LoRA

Fine-tune Qwen/Qwen3.5-4B on 95 (prompt, story) pairs so the model writes 300-word twists.

SFT · Instruction→
comic verse№ a-8

Limerick Spin LoRA

Fine-tune Qwen/Qwen3.5-4B on 70 (name, limerick) pairs so the model rhymes absurd five-liners.

SFT · Instruction→
story bible lore№ a-9

World Bible LoRA

Fine-tune Qwen/Qwen3-8B on 90 (faction, entry) pairs so the model writes consistent lore.

SFT · Instruction→
refrain poetry№ -10

Villanelle Weave LoRA

Fine-tune Qwen/Qwen3.5-4B on 55 (emotion, villanelle) pairs so the model knots refrains.

SFT · Instruction→
subtext dialogue№ -11

Dialogue Tag LoRA

Fine-tune Qwen/Qwen3-8B on 100 (beat, exchange) pairs so the model writes loaded talk.

SFT · Instruction→
concise revision№ -12

Prose Trim LoRA

Fine-tune Qwen/Qwen3.5-4B on 80 (flabby, tight) pairs so the model cuts dead weight.

SFT · Instruction→
mythic reframing№ -13

Myth Retell LoRA

Fine-tune Qwen/Qwen3-8B on 85 (myth, retelling) pairs so the model rewrites old gods.

SFT · Instruction→
end-word form№ -14

Sestina Loop LoRA

Fine-tune Qwen/Qwen3.5-9B on 50 (seed, sestina) pairs so the model rotates six words.

SFT · Instruction→
dramatic monologue№ -15

Monologue Heat LoRA

Fine-tune Qwen/Qwen3-8B on 75 (character, speech) pairs so the model voices inner storms.

SFT · Instruction→
erasure poetry№ -16

Found Poem LoRA

Fine-tune Qwen/Qwen3.5-4B on 65 (source, poem) pairs so the model carves found verse.

SFT · Instruction→
hip-hop verse№ -17

Slant Rhyme LoRA

Fine-tune Qwen/Qwen3-8B on 90 (topic, verse) pairs so the model writes near-rhyme bars.

SFT · Instruction→
romcom beats№ -18

Couch Scene LoRA

Fine-tune Qwen/Qwen3-8B on 70 (setup, scene) pairs so the model writes meet-cutes.

SFT · Instruction→
atmospheric description№ -19

Prose Mood LoRA

Fine-tune Qwen/Qwen3.5-4B on 85 (place, passage) pairs so the model sets gloomy tone.

SFT · Instruction→
interlocking verse№ -20

Pantoum Mirror LoRA

Fine-tune Qwen/Qwen3.5-9B on 45 (line, pantoum) pairs so the model echoes quatrains.

SFT · Instruction→
high-concept pitching№ -21

Logline Spark LoRA

Fine-tune Qwen/Qwen3.5-4B on 100 (idea, logline) pairs so the model compresses premises.

SFT · Instruction→
didactic short tale№ -22

Fable Moral LoRA

Fine-tune Qwen/Qwen3-8B on 80 (vice, fable) pairs so the model closes with a lesson.

SFT · Instruction→
dream poetry№ -23

Nocturne Line LoRA

Fine-tune Qwen/Qwen3.5-4B on 60 (dream, poem) pairs so the model writes sleepy imagery.

SFT · Instruction→
serial suspense№ -24

Cliffhang Curl LoRA

Fine-tune Qwen/Qwen3-8B on 95 (chapter, ending) pairs so the model snaps shut episodes.

SFT · Instruction→
song structure charts№ r-0

Beat Chart Weaver

Fine-tune Qwen/Qwen3-8B on 60 (prompt, ChordPro) pairs so the model outputs a valid ChordPro sheet from a plain-language brief every time.

SFT · Structured Output→
fixed poetic forms№ e-1

Sonnet Schema Forge

Fine-tune Qwen/Qwen3.5-4B on 80 (theme, Shakespearean sonnet JSON) pairs so it emits valid metered sonnet structures on demand.

SFT · Structured Output→
screenplay formatting№ r-2

Screenplay Block Builder

Fine-tune Qwen/Qwen3-8B on 70 (logline, Fountain script) pairs so it produces correctly tagged screenplay scenes from briefs.

SFT · Structured Output→
micro-poetry forms№ r-3

Haiku Lattice Coder

Fine-tune Qwen/Qwen3.5-4B on 100 (image cue, haiku JSON) pairs so it returns syllable-checked 5-7-5 poems as structured data.

SFT · Structured Output→
narrative skeleton DSL№ r-4

Myth Arc Mapper

Fine-tune Qwen/Qwen3-8B on 50 (premise, myth-arc DSL) pairs so it outputs a valid hero-journey graph from a story seed.

SFT · Structured Output→
rhyme pattern tables№ g-5

Rhyme Scheme Rig

Fine-tune Qwen/Qwen3.5-4B on 90 (mood, rhyme-scheme JSON) pairs so it emits valid stanza rhyme maps from a mood tag.

SFT · Structured Output→
visual narrative beats№ r-6

Storyboard Beat Cutter

Fine-tune Qwen/Qwen3-8B on 65 (scene brief, storyboard JSON) pairs so it returns shot lists with labeled beats every run.

SFT · Structured Output→
praise-poem structure№ y-7

Ode Form Foundry

Fine-tune Qwen/Qwen3.5-4B on 75 (subject, ode JSON) pairs so it outputs stanza-valid odes with address and refrain fields.

SFT · Structured Output→
script dialogue markup№ r-8

Dialogue Tag Tuner

Fine-tune Qwen/Qwen3-8B on 85 (exchange, tagged-dialogue JSON) pairs so it emits speaker-labeled lines with beat tags.

SFT · Structured Output→
complex rhyme forms№ r-9

Villanelle Vectorizer

Fine-tune Qwen/Qwen3.5-9B on 40 (theme, villanelle JSON) pairs so it outputs the 19-line form with repeated refrains correctly.

SFT · Structured Output→
book outline schema№ -10

Chapter Outline Oracle

Fine-tune Qwen/Qwen3-8B on 55 (synopsis, chapter-outline JSON) pairs so it returns numbered chapters with arc labels.

SFT · Structured Output→
comic verse form№ -11

Limerick Logic Loom

Fine-tune Qwen/Qwen3.5-4B on 95 (setup, limerick JSON) pairs so it emits anapestic 5-line jokes with AABBA rhyme.

SFT · Structured Output→
moral tale structure№ -12

Fable Frame Forger

Fine-tune Qwen/Qwen3-8B on 60 (animal premise, fable JSON) pairs so it outputs characters, lesson, and plot as fixed schema.

SFT · Structured Output→
invocation lists№ -13

Epic Catalog Coder

Fine-tune Qwen/Qwen3.5-9B on 45 (deity set, epic-catalog JSON) pairs so it returns enumerated invocations in strict order.

SFT · Structured Output→
dramatic solo speech№ -14

Monologue Mask Maker

Fine-tune Qwen/Qwen3-8B on 70 (character, monologue JSON) pairs so it emits beat-tagged solo speeches with emotion fields.

SFT · Structured Output→
word-repeat forms№ -15

Sestina Spin Setter

Fine-tune Qwen/Qwen3.5-9B on 35 (theme, sestina JSON) pairs so it outputs six stanzas with rotated end-words correctly.

SFT · Structured Output→
prose tone tags№ -16

Tone Row Typist

Fine-tune Qwen/Qwen3.5-4B on 100 (paragraph, tone-tag JSON) pairs so it labels sentences with mood and register fields.

SFT · Structured Output→
interlocking quatrains№ -17

Pantoum Pattern Press

Fine-tune Qwen/Qwen3.5-9B on 40 (memory, pantoum JSON) pairs so it returns mirrored quatrains with repeated lines.

SFT · Structured Output→
opening hook schema№ -18

Prologue Prompt Printer

Fine-tune Qwen/Qwen3-8B on 65 (world, prologue JSON) pairs so it emits hook, setting, and inciting field structured opens.

SFT · Structured Output→
coupled verse form№ -19

Ghazal Chain Grafter

Fine-tune Qwen/Qwen3.5-9B on 38 (beloved, ghazal JSON) pairs so it outputs rhymed couplets with refrained radif.

SFT · Structured Output→
serial beat endings№ -20

Cliffhang Coder

Fine-tune Qwen/Qwen3-8B on 80 (chapter text, cliffhang JSON) pairs so it returns ending type and tension tags.

SFT · Structured Output→
narrative song form№ -21

Ballad Stanza Binder

Fine-tune Qwen/Qwen3.5-4B on 90 (tale, ballad JSON) pairs so it emits ABCB stanzas with story-step fields.

SFT · Structured Output→
visual poem layout№ -22

Concrete Shape Scrivener

Fine-tune Qwen/Qwen3.5-4B on 85 (concept, concrete-poem JSON) pairs so it returns grid coordinates and word placements.

SFT · Structured Output→
letter-sequence novels№ -23

Epistolary Form Framer

Fine-tune Qwen/Qwen3-8B on 60 (plot, epistle JSON) pairs so it outputs dated letters with writer and arc fields.

SFT · Structured Output→
interlocked tercets№ -24

Terza Rima Rig

Fine-tune Qwen/Qwen3.5-9B on 42 (journey, terza-rima JSON) pairs so it emits ABA BCB CDC chained rhyme stanzas.

SFT · Structured Output→
constrained rhyme schemes№ r-0

Rhyme Lattice Forger

Fine-tune Qwen/Qwen3-8B on 60 couplet prompts so it emits verses passing a Python scansion-and-rhyme grader.

RL · Programmatic Reward→
metrical foot control№ r-1

Trochee Tuner

Fine-tune Qwen/Qwen3.5-4B on 40 line prompts so it outputs trochaic lines validated by a syllable-stress grader.

RL · Programmatic Reward→
syllable-limited forms№ y-2

Haiku Boundary

Fine-tune Qwen/Qwen3.5-4B on 50 haiku seeds so generations satisfy a 5-7-5 syllable checker.

RL · Programmatic Reward→
Shakespearean structure№ e-3

Sonnet Cage

Fine-tune Qwen/Qwen3-8B on 30 sonnet scaffolds so outputs pass rhyme-scheme and volta graders.

RL · Programmatic Reward→
screenplay beats№ t-4

Beat Drop Architect

Fine-tune Qwen/Qwen3-8B on 45 logline sets so it emits beats matching a structure-validator script.

RL · Programmatic Reward→
character voice№ k-5

Dialogue Spark

Fine-tune Qwen/Qwen3.5-4B on 70 dialogue seeds so lines pass a vocabulary-divergence grader.

RL · Programmatic Reward→
figurative language№ r-6

Metaphor Miner

Fine-tune Qwen/Qwen3-8B on 55 noun pairs so it yields metaphors scored by a novelty-distance grader.

RL · Programmatic Reward→
narrative arc shape№ r-7

Arc Bender

Fine-tune Qwen/Qwen3-8B on 35 outline prompts so plots satisfy a Freytag-pyramid checker.

RL · Programmatic Reward→
sound patterning№ h-8

Alliterance Smith

Fine-tune Qwen/Qwen3.5-4B on 50 phrase seeds so outputs maximise alliteration via a phoneme grader.

RL · Programmatic Reward→
poem segmentation№ r-9

Stanza Splitter

Fine-tune Qwen/Qwen3.5-4B on 40 raw poems so it inserts breaks passing a line-density grader.

RL · Programmatic Reward→
pacing curves№ -10

Tension Ramp

Fine-tune Qwen/Qwen3-8B on 45 scene sets so tension scores rise per a lexical-arousal grader.

RL · Programmatic Reward→
comic verse form№ -11

Limerick Loop

Fine-tune Qwen/Qwen3.5-4B on 35 limerick opens so closers pass AABBA rhyme and bounce graders.

RL · Programmatic Reward→
motif recurrence№ -12

Symbol Weave

Fine-tune Qwen/Qwen3-8B on 50 chapter seeds so symbol density matches a recurrence grader.

RL · Programmatic Reward→
authorial persona№ -13

Voice Mask

Fine-tune Qwen/Qwen3.5-4B on 60 style samples so outputs fool a perplexity-based authorship grader.

RL · Programmatic Reward→
chapter endings№ -14

Cliff Hang

Fine-tune Qwen/Qwen3-8B on 40 chapter bodies so ends maximise suspense via an open-loop grader.

RL · Programmatic Reward→
imagery specificity№ -15

Concrete Cut

Fine-tune Qwen/Qwen3.5-4B on 55 prompts so lines minimise abstractness via a noun-ratio grader.

RL · Programmatic Reward→
refrain repetition№ -16

Echo Chamber

Fine-tune Qwen/Qwen3.5-4B on 45 song frames so refrains match an exact-string grader.

RL · Programmatic Reward→
screenplay hooks№ -17

Cold Open

Fine-tune Qwen/Qwen3-8B on 35 scripts so cold opens pass an inciting-event grader.

RL · Programmatic Reward→
sentence rhythm№ -18

Prose Pulse

Fine-tune Qwen/Qwen3.5-4B on 50 paragraphs so length variance hits a coded rhythm grader.

RL · Programmatic Reward→
setting consistency№ -19

World Seed

Fine-tune Qwen/Qwen3-8B on 40 lore bits so expansions pass a fact-graph grader.

RL · Programmatic Reward→
wordplay generation№ -20

Pun Forge

Fine-tune Qwen/Qwen3.5-4B on 60 setups so puns score via a homophone-overlap grader.

RL · Programmatic Reward→
poem transitions№ -21

Verse Bridge

Fine-tune Qwen/Qwen3-8B on 45 stanza pairs so bridges pass a semantic-shift grader.

RL · Programmatic Reward→
thematic clarity№ -22

Theme Anchor

Fine-tune Qwen/Qwen3.8B on 50 drafts so theme tokens align with a keyword-coverage grader.

RL · Programmatic Reward→
microfiction№ -23

Flash Frame

Fine-tune Qwen/Qwen3.5-4B on 55 prompts so 100-word stories pass a twist-detector grader.

RL · Programmatic Reward→
long-form structure№ -24

Epic Spine

Fine-tune Qwen/Qwen3-8B on 30 saga seeds so books pass a multi-arc grader.

RL · Programmatic Reward→
studio house voice№ r-0

House Voice Taster

Fine-tune Qwen/Qwen3.5-4B on 40 preference triples so any generation lands in the studio's house voice.

RL · Preference Pairs→
free verse prosody№ r-1

Poetic Line Breaker

Fine-tune Qwen/Qwen3.5-4B on 60 line-break pairs so lineation matches a poet's breath and tension.

RL · Preference Pairs→
screenplay rhythm№ r-2

Screenbeat Pacer

Fine-tune Qwen/Qwen3-8B on 80 scene-pace pairs so act breaks hit trained dramatic timing.

RL · Preference Pairs→
figurative restraint№ r-3

Metaphor Pruner

Fine-tune Qwen/Qwen3.5-4B on 50 metaphor pairs so tropes stay fresh not cliché.

RL · Preference Pairs→
hook construction№ r-4

Cold Open Tuner

Fine-tune Qwen/Qwen3-8B on 70 opening pairs so cold opens grab without cheap tricks.

RL · Preference Pairs→
speech authenticity№ t-5

Dialogue Naturalist

Fine-tune Qwen/Qwen3.5-4B on 90 dialogue pairs so talk sounds lived-in not scripted.

RL · Preference Pairs→
sensitive tone№ r-6

Tone Safener

Fine-tune Qwen/Qwen3.5-4B on 60 tone pairs so hard topics keep respectful house posture.

RL · Preference Pairs→
songcraft rhyme№ r-7

Rhyme Schemer

Fine-tune Qwen/Qwen3.5-4B on 55 rhyme pairs so schemes feel witty not forced.

RL · Preference Pairs→
canon consistency№ r-8

World Lore Keeper

Fine-tune Qwen/Qwen3-8B on 75 lore pairs so expansions respect established canon voice.

RL · Preference Pairs→
art description№ r-9

Ekphrasis Painter

Fine-tune Qwen/Qwen3.5-4B on 50 ekphrasis pairs so object writing stays luminous not flat.

RL · Preference Pairs→
narrative structure№ -10

Beat Drop Curator

Fine-tune Qwen/Qwen3-8B on 80 beat pairs so story turns satisfy trained structural taste.

RL · Preference Pairs→
sentence ornament№ -11

Prose Decorator

Fine-tune Qwen/Qwen3.5-4B on 65 syntax pairs so decoration enhances not buries meaning.

RL · Preference Pairs→
interior voice№ -12

Monologue Whisperer

Fine-tune Qwen/Qwen3.5-4B on 55 monologue pairs so inner speech feels true not theatrical.

RL · Preference Pairs→
quiet scene№ -13

Slice of Life

Fine-tune Qwen/Qwen3.5-4B on 60 scene pairs so mundane moments stay textured not dull.

RL · Preference Pairs→
comic timing№ -14

Punchline Smith

Fine-tune Qwen/Qwen3.5-4B on 70 joke pairs so setups land with trained comic taste.

RL · Preference Pairs→
literary translation№ -15

Translation Echo

Fine-tune Qwen/Qwen3-8B on 75 translation pairs so renditions keep source literary flavor.

RL · Preference Pairs→
short form№ -16

Haiku Slimmer

Fine-tune Qwen/Qwen3.5-4B on 50 haiku pairs so cuts deepen image not thin it.

RL · Preference Pairs→
ending design№ -17

Arc Resolver

Fine-tune Qwen/Qwen3-8B on 80 ending pairs so closures satisfy trained resolution taste.

RL · Preference Pairs→
author emulation№ -18

Voice Mimic Safe

Fine-tune Qwen/Qwen3.5-4B on 65 emulation pairs so mimicry stays respectful not derivative.

RL · Preference Pairs→
implicit meaning№ -19

Subtext Weaver

Fine-tune Qwen/Qwen3-8B on 70 subtext pairs so layers imply without stating.

RL · Preference Pairs→
trope placement№ -20

Genre Boundary

Fine-tune Qwen/Qwen3.5-4B on 60 genre pairs so tropes sit fresh inside set conventions.

RL · Preference Pairs→
prose cadence№ -21

Rhythm Scanner

Fine-tune Qwen/Qwen3.5-4B on 55 cadence pairs so sentences breathe with trained rhythm.

RL · Preference Pairs→
pitch voice№ -22

Query Letter Polisher

Fine-tune Qwen/Qwen3.5-4B on 65 query pairs so letters charm agents with right tone.

RL · Preference Pairs→
moral closure№ -23

Fable Moralizer

Fine-tune Qwen/Qwen3.5-4B on 50 fable pairs so morals land earned not preached.

RL · Preference Pairs→
avant form№ -24

Experimental Bender

Fine-tune Qwen/Qwen3.5-9B on 85 form pairs so experiments stay coherent not random.

RL · Preference Pairs→