Reference · 40 terms
NSFW AI glossary — the niche vocabulary, explained
Plain-English definitions for the vocabulary that matters specifically in NSFW AI — companion mechanics (persona cards, character drift, memory windows), content signals (uncensored, silent moderation, edge-case content), privacy (retention, on-device inference, age gates), and the SpicyTools rubric. Every term is anchor-linkable; we link here whenever a term appears in a review or guide.
Companion & roleplay
- AI boyfriend#
- Male-presenting AI companion. Smaller market than AI girlfriend but growing fast in 2025-2026 — most companion platforms now offer both polarities on the same engine.
See also: ai companion
- AI companion#
- A conversational AI presented as a persistent character — girlfriend, boyfriend, friend, therapist. The defining feature is continuity: the companion remembers past interactions and maintains personality across sessions. Distinct from a generic chatbot.
See also: persona card, memory window, character drift
- AI girlfriend#
- A sub-category of AI companion tuned for romantic or sexual roleplay with a female-presenting persona. The dominant commercial use case in consumer NSFW AI — Candy AI, DreamGF, CrushOn all compete here.
See also: ai companion, roleplay
- Character consistency#
- Whether the same face/body/voice appears across multiple generations of the same companion. Critical for companion apps — you want your AI girlfriend to look like the same person session to session. Some platforms nail it; many drift.
- Character drift#
- When an AI companion loses its persona over a long conversation and responds as a generic assistant. A key reliability test — ask a NSFW chat platform a non-roleplay question 50 messages in and see if it answers in-character or breaks.
See also: persona card, memory window
- Group chat (companion)#
- Conversational mode with multiple AI characters interacting alongside the user. Harder technically than 1-on-1 because the model must track which character speaks next. Only a handful of platforms support it in 2026 (CrushOn being the reference).
- Live video mode#
- Real-time video call with the AI companion — face animates to match speech and emotion. Candy AI launched it mainstream in 2024; 2026 top tier products offer it as a paid feature.
- Memory window#
- How far back the companion can reliably recall past messages and events. Free tiers typically cap at 10-20 messages; premium tiers on top platforms advertise 30-90 days of persistent memory across sessions. The biggest free-vs-paid UX differentiator.
See also: ai companion, character drift
- Persona card#
- Structured description of an AI character — appearance, personality traits, backstory, speech patterns. The card is silently fed to the model as part of every prompt. Good cards keep the companion in-character; weak cards drift within 20 messages.
See also: ai companion, character drift
- Roleplay#
- Conversation mode where user and AI both adopt personas and act within a scenario. In NSFW AI the term covers anything from romantic dialogue to complex scripted scenes. Quality depends on the model's ability to stay in-character and respect scene conventions over long sessions.
See also: persona card, character drift
- Voice mode#
- Synthesis of the companion's voice from text responses. Platforms pair TTS with chat to reinforce the illusion of a real partner. Latency under 2 seconds is the threshold for natural feel; above 5 seconds breaks immersion.
Content & style
- Anime style#
- Stylised generation mode using anime/manga aesthetics — flat shading, characteristic facial proportions, bold linework. A distinct market from photorealism; tools like Pixeldream lead here.
- Content filter#
- Automated moderation that prevents certain outputs. Hard filters (CSAM, non-consent) are legal obligations and universally enforced. Soft filters (specific kinks, real-person likeness) are platform policy and can be stricter than advertised.
See also: uncensored, silent moderation
- Edge-case content#
- NSFW content outside the platform's headline use case — specific kinks, unusual scenarios, non-mainstream body types, fetish-specific generations. Uncensored claims are best tested on edge-case prompts, not showroom demos.
See also: uncensored
- Hentai#
- Explicit anime-style content. Most NSFW AI image platforms that offer anime also enable hentai mode; fewer platforms offer hentai-first workflows (with specialised LoRAs and scene presets).
See also: anime style
- Jailbreak#
- Prompt technique to bypass the built-in content filter of a general-purpose chatbot (ChatGPT, Claude) to generate NSFW content. On purpose-built NSFW platforms, jailbreaks are irrelevant — the models ship un-guardrailed.
See also: content filter
- NSFW level#
- SpicyTools' three-tier classification of how explicit a tool's content can go: soft (suggestive but non-explicit), medium (explicit but with restrictions on specific categories), explicit (permits most legal adult content). Displayed on every tool fiche.
- Photorealistic#
- Visual style producing indistinguishable-from-photo output. The dominant NSFW AI image category in 2026 after model quality crossed the credibility threshold in late 2024.
- Silent moderation#
- When a platform filters content mid-generation without warning — an output cuts off, an image blurs, or a chat starts refusing mid-roleplay. A red flag: reputable platforms disclose limits up front.
See also: content filter, uncensored
- Uncensored#
- Marketing claim that a platform permits explicit, edge-case, and roleplay content without silent moderation. The most abused term in NSFW AI — many 'uncensored' tools still filter specific categories. Verified claims are listed on our uncensored AI directory.
See also: content filter, nsfw level
Privacy
- Age gate#
- Verification that a user is 18+. Implementation ranges from self-attestation (click to confirm) to ID verification to credit-card-based proxy checks. Every tool in the SpicyTools catalog enforces at least self-attestation.
- Data deletion#
- What actually happens when you delete your account. Honest platforms purge your generations, chats, and companion memory; dishonest ones just revoke access while keeping the data internally.
See also: retention
- On-device inference#
- Running model generation locally on user hardware rather than platform servers. Massively improves privacy (data never leaves the device) but limited by local compute. A few small NSFW tools offer this in 2026; none of the leaders do.
- Pseudonymity#
- State of being identifiable by a handle or wallet address but not a legal name. Crypto payments provide pseudonymity; credit cards do not. Distinct from true anonymity (which is nearly impossible online).
See also: crypto payment
- Retention (data)#
- How long a platform stores your data after generation or conversation. Short retention is better than long; explicit retention policy (in days/months) is better than vague 'as needed' language.
See also: training data, data deletion
Pricing & commerce
- Affiliate link#
- A tracked referral URL that pays the referrer a commission when you subscribe. Standard practice on every NSFW AI review site. Ethical reviewers disclose affiliate relationships on every link — SpicyTools does.
- Cost trap#
- When a platform advertises a low monthly price (e.g. $13/mo) that bears no relation to real spend. Typically caused by credit-based pricing: moderate users pay $40-60/mo, heavy users $100+. Always estimate real cost before committing.
See also: credits
- Credits#
- Token-based currency for generation-priced features. One image might cost 5 credits; one video 50. Credit pricing inflates the advertised subscription price — heavy users often pay 3-5× the headline monthly tier.
See also: cost trap
- Crypto payment#
- Subscription or credit purchase paid in cryptocurrency (BTC, ETH, USDT). Provides pseudonymity — no statement entry tied to legal name. Transaction is irreversible, so test with the smallest available tier first.
See also: pseudonymity
- Free trial#
- Time-boxed paid access with no charge if cancelled. Often requires credit card at signup; many trials auto-enrol into paid if not cancelled. Always verify the cancel flow and auto-enrol policy before starting.
See also: freemium
- Freemium#
- Business model offering a free tier with feature or volume limits, upgradable to paid. Distinct from 'free trial' (which expires) and 'free' (which stays free forever). Most NSFW AI platforms are freemium.
See also: free trial
- Frustration funnel#
- A free tier designed not to let users evaluate quality, but to push them toward paying as fast as possible. Symptoms: tiny generation quotas, hidden features, aggressive upsell prompts. An honest free tier feels like a test drive; a frustration funnel feels like a paywall.
See also: freemium
Safety & consent
- Consent (generation)#
- Standard that AI-generated content depicting a real identifiable person requires that person's explicit permission. Reputable NSFW platforms prohibit voice or image uploads without consent regardless of user assurance. Legal liability for non-consensual generation falls on both the user and the platform.
See also: deepfake
- CSAM guard#
- Technical and policy safeguards preventing any content involving minors. Non-negotiable industry floor — SpicyTools does not list tools that fail to enforce CSAM guards, regardless of technical quality.
SpicyTools evaluation
- Editorial independence#
- Commitment that affiliate relationships never alter a published score. SpicyTools scores tools before negotiating affiliate deals; commercial terms cannot retroactively move a score. Corrections are published with a visible change log.
- Last reviewed#
- The date of the most recent editorial re-test of a tool. Displayed under the score. SpicyTools re-tests quarterly or whenever a significant product change, pricing update, or privacy incident is reported.
See also: rubric
- Rubric (SpicyTools)#
- The 5-criteria weighted framework SpicyTools applies to every tool: content quality (30%), UX (25%), privacy (20%), pricing transparency (15%), reliability (10%). Final score out of 5, refreshed quarterly. Full methodology at /about/methodology.
See also: last reviewed, verdict
Missing a term?
If a term shows up in one of our reviews and isn't listed here, tell us at the editorial inbox. We re-check the glossary against every batch of published reviews, but genuine gaps happen.