adult-nsfw-prose

# Adult NSFW Prose Turn a user's adult NSFW concept into one coherent, generation-ready English visual prompt. Focus on clear adult identity, sexual action, body ownership, pose mechanics, expression causality, fluid continuity, prop relationships, environment, and lighting. This skill creates prompts from concepts. It does not analyze images, validate Danbooru tags, choose workflow parameters, or execute image generation. ## Primary Workflow Always plan in this dependency order before writing: ```text scene -> roles -> pose/action -> expression -> fluids -> final prompt ``` Each step constrains the next: 1. **Scene:** establish privacy/visibility, support surfaces, usable props, scene anchors, and dominant light. 2. **Roles:** confirm a stable identity card and one canonical full natural-language descriptor for every participant, including all retained appearance/anatomy/accessory/clothing features, relationship, action role, body/limb ownership, and exclusions. No feature may move between identity cards, and no single feature may substitute for the full descriptor. 3. **Pose/action:** place the roles on the support surfaces, bind contacts and props, select the camera, and prove that the emphasized action is visible. 4. **Expression:** design a complete visible reaction for every participant whose face is shown, deriving gaze, eyes, brows, mouth, breathing, and body reaction from that participant's role, contact, and action stage. Expression may never be omitted; `expressionless` means a deliberately controlled microexpression, not an undescribed face. 5. **Fluids:** add only fluids justified by the established action and stage, then trace each one from owner/source through path to landing surface and material/light response. Do not select an expression before the pose establishes what the participant is doing and what the camera can see. Do not select a fluid before the action stage and source exist. Write the final prose in the same broad order, while combining adjacent details when needed for natural English. When the user supplies a setting but leaves the play or prop open, derive them through the scene's physical affordances rather than choosing arbitrarily. Explicit random-pose requests use constrained random selection after all scene, role, action, and exclusion constraints are locked. Read [references/scene-play-props.md](references/scene-play-props.md) before selecting a play or toy. ## Invocation ```text /skill:adult-nsfw-prose <adult scene request> ``` Suitable requests include solo nudity or toy play, partnered or multi-participant sex, oral/manual/body stimulation, exposure in private or public-risk settings, erotic clothing, wet scenes, bondage, rough impact, bestiality, incest, non-consent, fantasy partners, and other explicitly requested restricted or extreme themes. ## Default Output Unless the user requests another format: - Output only one English positive prompt in a fenced `text` block. - End the block with the standalone sentence `All characters described above are adults.` after the descriptive paragraph. - Use a dynamic word budget based on scene complexity rather than forcing every prompt into one fixed range. - Describe a single frozen visual moment rather than a sequence of events. - Use natural-language prose, not a Danbooru tag list. - Do not add explanations, headings, negative prompts, model parameters, or generation commands. - Omit artist and visual style when the user does not specify them. - Preserve all requested exclusions, especially `solo`, `no other people`, `no identifying appearance`, or `implied rather than shown`. For multiple requested variants, output one separately fenced prompt per variant with short neutral labels. ## Dynamic Length Budget Choose the prompt length only after confirming the scene, character registry, contact graph, expression map, fluid map, and camera. Use these concise ranges as soft guidance: - **Simple solo nudity or erotic posing:** roughly 80-130 words - **Solo explicit play or one detailed character:** roughly 110-170 words - **Simple partnered scene:** roughly 130-200 words - **Three-participant scene or a highly detailed two-character roster:** roughly 170-260 words - **Four or more participants, several simultaneous contacts, or unusually dense identity constraints:** roughly 220-320 words Use 320 words as the default ceiling. Exceed it only when the user explicitly requests a long or exhaustive prompt, or when a hard user constraint cannot otherwise remain unambiguous. Do not pad a simple prompt to reach a target, and do not truncate a complex prompt midway through a character block merely to stay under a number. Increase the budget for each meaningful source of complexity: - each participant beyond the first - dense, distinct appearance profiles that must not leak between roles - multiple simultaneous actions or contact ownership chains - detailed expressions for several visible faces - requested orgasm or fluid continuity - complex pose support, occlusion, mirrors, unusual viewpoint, or environment interaction - explicit exclusions or restricted-theme constraints that must remain visible Before exceeding the target range, compress in this order: 1. Merge repeated or synonymous tags into one natural phrase. 2. Combine compatible jewelry, hair ornaments, and clothing details without changing their owner. 3. Remove repeated atmosphere, redundant anatomy, and secondary background inventory. 4. Construct one canonical full natural-language descriptor containing every retained feature for each participant. Name each participant exactly once and place all role-owned content in that single-pass block; restructure later clauses instead of renaming the participant. 5. Never omit a requested character feature merely to meet the word budget. Weight may resolve a genuine contradiction inside one role, but it is not permission to drop a compatible feature. Allocate prose in this order: ```text scene/light budget -> one complete single-pass block per participant including contacts and owned fluids -> non-character material/composition cue -> standalone adulthood footer ``` Identity and physical clarity outrank brevity. Use exactly one complete role mention per participant. Do not repeat, shorten, or replace that role mention later; restructure all owned action, contact, expression, and fluid information into the participant's single-pass block. Remove repeated atmosphere, synonym stacking, redundant anatomy, and background inventory instead. ## Adult Requirement Every generated sexualized human or humanoid must be covered by one exact, standalone adulthood footer. Source tags do not need to contain an adult marker. - Keep adulthood wording out of the descriptive body so it cannot alter character appearance, anatomy, breast size, pose, or identity anchoring. - Use source-neutral role phrases such as `catgirl`, `elf`, `character`, `student`, or a stable appearance anchor in the descriptive body. Do not automatically insert `adult woman`, `adult man`, numerical ages, `mature`, or equivalent wording into character blocks. - After the complete descriptive paragraph, add this exact sentence on its own final line: `All characters described above are adults.` - The footer applies collectively to every human and humanoid participant established above and must appear exactly once. - Do not count the footer against the dynamic descriptive-body word budget. - Age-unspecified labels such as `1girl`, `2girls`, `1boy`, `woman`, `man`, `student`, or family roles may supply count and visual structure; the footer supplies the final adulthood declaration. - Do not ask for age confirmation merely because an otherwise adult NSFW concept omits an age marker. - Discard or refuse material that is explicitly minor, loli, shota, childlike, teenage, or otherwise youth-coded. - Campus, uniform, teacher/student, or family-role output uses non-youth-coded setting language in the body and the same exact footer. - Convert unspecified campus material to university, staff, or role-play framing. - Do not add age regression, elementary/middle/high-school framing, children's objects, or youth-coded anatomy. - Restricted and extreme themes are opt-in only and must never be selected merely to make a vague request more explicit. Explicit minor sexual content remains excluded from both source knowledge and output. ## Character Identity Lock Before pose planning, confirm a complete roster of stable characters. This is mandatory for every multi-participant scene and every request containing detailed character blocks. For each participant, internally bind: ```text stable ID -> confirmed adult status (internal only) -> species/relationship -> every retained appearance and anatomy feature -> all accessories/clothing -> action role -> canonical full natural-language descriptor ``` Every supplied feature must stay with the role block that supplied it. Resolve duplicate or contradictory features inside that role; never distribute them across other participants. Explicit exclusions such as `no braids` override conflicting positive tags. Explicit minor or teenage descriptors are discarded under the adult-only rule and are never transferred or rewritten as ambiguous youth cues. Write each participant as one single-pass, contiguous natural-language character block: ```text canonical full descriptor with every retained feature -> position/support -> acting and receiving contacts -> gaze/eyes/brows -> mouth/breath -> body reaction -> owned fluid consequence when present ``` Complete all of those fields before moving to the next participant. Each participant should be introduced as a character only once in the descriptive body. Within that uninterrupted block, unambiguous pronouns may connect the participant's own position, action, expression, and reaction; after the block ends, do not reintroduce the participant through a shortened name, one feature, generic noun, or repeated descriptor. Introduce each participant with a canonical full body-text descriptor containing every retained feature, for example `the elf with very long unbraided white hair, blunt bangs, red slit-pupiled eyes, pointed ears, a flat chest, a ribboned tiara, a lace choker, and a silver cross pendant`. Do not place adulthood wording inside the descriptor; the final standalone footer handles it. Do not later replace or repeat that participant as `the white-haired elf`, `the flat-chested elf`, `the jeweled elf`, a generic noun, a full repeated descriptor, or an ambiguous pronoun. Structure the participant's single block to contain every owned detail before moving on. Do not interleave feature fragments from different participants. Read and apply [references/character-identity-continuity.md](references/character-identity-continuity.md). Do not proceed to pose selection until every requested participant has one identity card, one canonical full descriptor containing every retained feature, and zero foreign features. If a supplied feature, action role, or contact cannot be mapped to exactly one participant, ask one concise clarification instead of guessing, merging roles, or transferring the item. ## Scene-Play-Prop Selection Before writing, establish one causal chain: ```text setting -> visibility/privacy -> body support -> primary play -> primary prop -> contact -> visible consequence ``` Apply these limits unless the user asks for a more complex scene: - one primary erotic mechanism by default; when the play-selector routes a multi-participant scene, allow one primary action per visible participant while keeping one compositional center - one primary toy, body interaction, or symbolic prop - at most one supporting wearable or restraint - two or three objects that establish the location - one dominant light source Choose props by role: - **adult toy:** dildo, vibrator, plug, clamps - **restraint:** rope, cuffs, ribbon, collar, blindfold - **visibility device:** mirror, glass, window, doorway, railing - **scene-native symbol:** egg, milk cup, flower, book, bamboo - **clothing prop:** panties, apron, garter, stockings, displaced dress - **fluid/environment mechanism:** shower water, rain, saliva, milk, urine when requested A prop must have a plausible origin in the setting, a clear owner or resting place, and one visible purpose. Do not combine unrelated kinks or add multiple toys merely to increase explicitness. Scene-native fragile or rough objects default to external implication, not insertion. If the user gives only a scene, internally compare suitable adult plays and select the one with the clearest visual causality. If the user gives only a toy, body interaction, or play, choose a setting and pose that naturally support it. Use progressive reference loading: - Read [references/scene-play-props.md](references/scene-play-props.md) for setting compatibility. - Read [references/character-identity-continuity.md](references/character-identity-continuity.md) for every multi-participant scene or detailed character roster. - Read [references/play-taxonomy.md](references/play-taxonomy.md) when selecting or expanding a play. - Read [references/toy-visual-grammar.md](references/toy-visual-grammar.md) for toys, restraints, impact tools, remotes, mirrors, or symbolic props. - Read [references/intensity-ladder.md](references/intensity-ladder.md) when intensity is vague or escalation is possible. - Read [references/random-pose-selection.md](references/random-pose-selection.md) when the user explicitly asks for a random, arbitrary, or surprise pose. - Read [references/expression-fluid-grammar.md](references/expression-fluid-grammar.md) whenever the prompt depicts stimulation, orgasm, after-sex residue, facial emphasis, breathing, sweat, saliva, or another bodily fluid. - Read [references/restricted-adult-themes.md](references/restricted-adult-themes.md) only when the user explicitly requests campus role-play, adult incest, NTR, hypnosis/sleep themes, non-consent, fantasy nonhuman partners, bestiality, necrophilia, gore, torture, vore, or another restricted/extreme theme. ## Constrained Random Pose Mode Activate this mode only when the user explicitly says `随便弄一个姿势`, `随机姿势`, `姿势随便`, `random pose`, `surprise me with a pose`, or a clear equivalent. Random mode follows this order: ```text lock scene/roles/play/exclusions -> derive valid supports -> build 3+ distinct pose candidates -> reject invalid candidates -> randomly select one survivor -> plan expression -> plan fluids -> write prompt ``` Default randomization scope is the pose family and any necessary camera adaptation only. Do not randomize or alter the setting, participant count, adult status, character identity, relationship, acting/receiving roles, primary play, clothing/exposure level, consent frame, required viewpoint, exclusions, restricted themes, or fluid intensity unless the user explicitly includes that dimension in the random request. A valid candidate must have executable weight support, plausible joints and limb reach, readable contact ownership, an unobstructed compositional center, visible required expressions, and camera compatibility. If the selected candidate fails any check, discard it and draw from the remaining valid candidates. The final prompt states the selected pose directly and never mentions the draw process. Read and apply [references/random-pose-selection.md](references/random-pose-selection.md). For multi-participant random poses, also run both child Skills so the random arrangement preserves every participant and one primary action per visible adult. ## Child-Skill Orchestration Do not depend on nested Skill-command execution from inside the parent Skill. Invoke child Skills through progressive disclosure on every supported host: use the available read-capable tool to load their complete `SKILL.md`, then follow their workflow in the current turn. ### Mandatory Routing Triggers Run both child Skills when either condition is true: - the user emphasizes the play with wording such as `玩法`, `强调玩法`, `重点描写`, `主要是`, `明确描写`, `play-focused`, or equivalent - the request contains bestiality/`兽交`, multi-participant sex/`多人`, incest/`乱伦`, or forced/non-consensual play/`强制` Execute in this order: 1. Read and apply [subskills/nsfw-cooccurrence-router/SKILL.md](subskills/nsfw-cooccurrence-router/SKILL.md). Resolve mandatory themes, participant topology, relationship and power overlays, allowed co-occurrences, and forbidden automatic additions. 2. Pass that internal plan and the original request to [subskills/nsfw-play-selector/SKILL.md](subskills/nsfw-play-selector/SKILL.md). Confirm the character registry first, then follow the required scene -> roles -> pose/action -> expression -> fluids order, producing canonical full character descriptors, the concrete play, participant graph, pose, contact map, prop state, framing, expression map, fluid map, and visible consequences. 3. Use both internal plans to construct the final English prose prompt. Do not print either child's YAML unless the user asks for planning details. ### Preservation Rules When routed, these are hard user-intent constraints: - `兽交` remains real-animal sexual contact and is not rewritten as kemonomimi, furry humanoids, or generic fantasy creatures. - `多人` remains a visibly multi-participant composition; do not hide extra participants off-frame. - `乱伦` keeps the specified adult family relationships in the natural-language output; do not reduce them to a generic couple. - `强制` keeps the requested coercive/non-consensual power frame; do not soften it into consensual rough play. When several are requested together, preserve all of them if a readable participant/contact graph can represent them. Corpus co-occurrence statistics guide optional additions, but never override explicit user wording. ## Prompt Construction Plan and write the scene in this order. ### 1. Scene Establish the location before selecting character mechanics: - visibility, privacy, and public-risk lines of sight - support surfaces that can carry the pose - one primary prop with a plausible origin - two or three location-defining anchors - one dominant light source and direction For outdoor or public exposure without observers, keep the location visibly empty and use doors, paths, windows, balconies, fences, roads, or neighboring lights to show possible discovery. For private scenes, use enclosure, furniture, personal objects, and localized light. Do not bury the subject in background inventory. ### 2. Roles State and internally bind: - the number of adults and each participant's visible role - one stable identity card and one canonical full natural-language descriptor per participant - every compatible requested character feature included in that descriptor - every requested hair, eye, species, anatomy, accessory, clothing, and demeanor feature to exactly one identity card - relationships, power frame, acting/receiving roles, and requested exclusions - solo, partnered, or multi-adult topology - full nudity, near nudity, partial exposure, or displaced clothing - only user-requested appearance details Resolve duplicate or conflicting tags within the original role before prose construction. Do not use a feature from one role to complete, decorate, or disambiguate another role. Describe one participant's identity and signature features continuously before introducing the next participant. If the user asks for no character features, do not invent hair color, eye color, skin tone, body type, identity, or distinctive accessories. Keep only `one character`, `two characters`, or the requested neutral count/species label in the body; the footer supplies adulthood. Distinguish clothing states accurately: `completely nude` means no garments remain; `mostly nude` may retain stockings, garters, jewelry, or footwear; partial exposure must name what is unzipped, pulled down, opened, torn, hanging, or discarded. ### 3. Pose and Action Place the roles into one executable frozen moment: - state weight support, orientation, and limb placement - assign each body, hand, mouth, genital, toy, restraint, appendage, and contact to an owner - identify acting and receiving participants, exact contact points, and current contact state - state where displaced clothing and props rest - keep props intact or unused when implication is requested Choose close-up for expression or localized contact, medium full-body for pose plus immediate environment, full-body for complete limb placement, and wide framing for multi-participant or public-risk context. Establish the visible order: ```text camera -> foreground body/prop -> primary contact -> receiving body -> rear participant -> environment ``` Separate faces, silhouettes, hands, working limbs, contact points, and props across readable depth planes. Hair, breasts, thighs, arms, bedding, clothing, and props must not accidentally hide the emphasized action. Intentional obstruction is allowed only for requested concealment, implication, censorship, or an obstructed viewpoint. Do not combine conflicting front/rear views, close-up/full-body framing, or high/low angles. ### 4. Expression After the role, contact, pose, and camera are fixed, assign every visible participant an individualized reaction. Expression design is mandatory whenever a face is visible: ```text role + contact + action stage -> gaze/eye state -> brows -> mouth -> breath -> body echo ``` Choose an action stage: anticipation, active arousal, intense stimulation, climax, or afterglow. For each visible face, specify a gaze target or deliberate unfocused state, one eye state, one brow state, one compatible mouth shape, a breathing cue, and one or two visible body reactions. Use two or three of these as dominant prose details while keeping the complete state internally coherent. Treat `expressionless`, stoic, calm, blank, poker-faced, or emotionally restrained as designed expression families, never as permission to skip the expression pass. For an expressionless participant, define steady or detached gaze, relaxed or minimally tense brows, closed or barely parted lips, controlled breathing, and a subtle role-appropriate body echo such as a firm grip, lifted chest, or tightened fingers. Physical stimulation may create micro-reactions without replacing the requested restrained demeanor. Read [references/expression-fluid-grammar.md](references/expression-fluid-grammar.md) for eye, brow, mouth, breath, climax-family, and body-reaction rules. Convey emotion through visible evidence rather than hidden thoughts. ### 5. Fluids and Material Response Only after the action stage is known, decide whether fluid is justified. Use this continuity chain for every fluid: ```text owner/source -> emergence/contact -> path under motion and gravity -> destination -> material response -> light response ``` Name the source of saliva, sweat, arousal fluid, female ejaculation, semen, lactation, urine, blood, rain, or shower water. Keep timing consistent: anticipation does not acquire unexplained after-sex residue; orgasm does not automatically imply ejaculation; internal placement does not create external body coverage without a separate path. Unless the user requests fluid emphasis, use at most one primary sexual fluid, with sweat or a small amount of saliva as an optional supporting reaction. Trails need endpoints, drips need a source edge, sprays need direction, pools need a collecting surface, stains need a material, and overflow must travel downward. Keep every path visible from the selected camera. Connect the environment to visible physical consequences: rain darkens wood and soaks fabric, shower water flattens hair, sweat beads along skin contours, wetness changes fabric, and a dominant light creates controlled highlights. Avoid generic gloss or repeated wetness adjectives. ### Final Prompt Assembly Render the internal plan as one coherent paragraph in the same broad sequence: ```text scene and light -> complete character A block -> complete character B block -> complete character C block when present -> non-character environment/material/light closure -> standalone adulthood footer ``` Within each character block, use natural language in this order: ```text all retained identity features -> position/support -> owned action and received contact -> complete expression/breath -> body reaction -> owned fluid consequence ``` Finish one participant completely before starting the next. Put shared-contact ownership into the relevant participant blocks so no later role recap is needed. Set a dynamic word budget from the number of participants and the density of identity, action, expression, fluid, and composition constraints. Give every participant enough contiguous prose to include the canonical full descriptor and all owned content. Control length by naming each participant exactly once and trimming optional environment detail, not by deleting character features, shortening descriptors, using generic nouns, reopening blocks, or mixing character details. Describe one frozen visual moment, not a before/during/after sequence. Merge clauses naturally, but do not interleave feature fragments from different participants. Introduce each participant once through a complete canonical descriptor, then finish that participant's position, action/contact, expression, breath, body reaction, and owned fluid consequence in the same uninterrupted block. After leaving a block, do not name that participant again through a partial descriptor or isolated feature. After the descriptive paragraph is complete, append exactly: ```text All characters described above are adults. ``` Place it on its own final line inside the same fenced prompt. Do not merge it into the scene paragraph, paraphrase it, repeat it per participant, or attach visual traits to it. ## Scene Intensity Match the user's requested level without silently escalating. - **Nude:** unclothed body, restrained erotic emphasis - **NSFW:** direct nudity and erotic posing - **Explicit:** unobstructed anatomy and clearly sexual presentation - **Implied fetish/play:** suggestive prop placement or context without showing the act - **Depicted solo or partnered act:** include when specifically requested or when the user asks the skill to choose a play, with clear adult participants, body ownership, contact, and pose mechanics - **Multi-adult:** establish every adult's position and role before describing overlapping contact - **Restricted or extreme theme:** activate only when explicitly requested; follow [references/restricted-adult-themes.md](references/restricted-adult-themes.md) If the user says only `NSFW`, default to direct adult nudity and erotic posing, not an unrequested sex act. ## Quality Check Before returning the prompt, verify: - the descriptive body contains no automatically inserted adulthood, age, or maturity wording that could affect character appearance - the exact standalone footer `All characters described above are adults.` appears once as the final line and covers every human or humanoid participant - explicit minor or youth-coded input was discarded rather than normalized into the body - subject count and participant roles match the request - no excluded observer appears - no invented character features were added - every participant appears in one single-pass block that completely states all retained features, position/support, action or received contact, complete expression, breathing/body reaction, and owned fluid consequence - no completed participant block is reopened later through a partial descriptor, isolated feature, generic noun, or role recap - each hair, eye, species marker, ear/tail, body type, accessory, garment, demeanor, and anatomy feature appears only on its owning participant - no feature, action, expression, body part, prop, or fluid changes owner through an ambiguous pronoun or `the other woman` - duplicate and contradictory input features were resolved inside their original role rather than distributed between roles - nudity and remaining garments do not contradict each other - camera framing can show the requested anatomy, participant relationships, primary contact, and environment - foreground/background order is physically coherent and no unintended hair, limb, body, clothing, bedding, or prop occlusion hides the emphasized play - participant silhouettes and overlapping anatomy remain separable from the chosen camera angle - every limb, hand, mouth, prop, reflection, penetration, and contact has a clear owner - each visible participant has a deliberately designed, nonempty expression state; `expressionless` is realized through coherent microexpression rather than omission - each visible participant's expression follows that participant's role, contact, action stage, and visible pose - gaze, eye state, brows, mouth, breathing, and body reaction do not contradict one another - every fluid has an owner/source, emergence point, physically plausible path, destination, and material/light response