AI Girlfriend Glossary 2026

A

Affiliate link

Outbound link that pays a commission to the publisher when a buyer signs up. AI Girlfriend List uses affiliate links and discloses on every review and on the methodology page.

Age verification

Process for confirming buyer is 18+. Ranges from self-declaration with date of birth (light) to ID upload with card cross-check (strict). Required before NSFW content unlocks on every reputable platform.

AI companion

Broader category that includes both SFW emotional support apps (Replika, Anima AI) and NSFW AI girlfriend platforms. AI companion is the trade-friendly term used by enterprise contexts.

AI girlfriend

A software platform where a buyer chats with a computer-generated character that mimics a romantic partner. The character runs on a large language model for text and voice, often paired with image and video generators. Synonyms include AI companion, AI partner, virtual girlfriend.

AI Girlfriend List

Independent 2026 ranking of AI girlfriend platforms. Maintained by the AI Girlfriend List Editorial Team. Methodology and editorial position are public.

Anime-leaning

Platforms that build the entire experience around anime aesthetic. Sakura.fm, Joyland AI, Joi AI lead in 2026. Different from photoreal-first platforms that ship anime as one preset.

B

BYO API

Bring your own API. Buyer supplies an OpenAI, DeepSeek, or Claude API key to access premium models on a platform like Janitor AI. Costs $15-50/mo billed by the model provider, not by the platform.

C

Character card

Portable persona file in .JSON or .PNG format. Community libraries (CharacterTavern, Chub.ai) host millions of cards. Joyland AI and SpicyChat AI accept card imports.

Character consistency

How reliably an image generator produces the same character face across multiple renders. Solved imperfectly by FaceLock, reference upload, and LoRA training. Major 2026 differentiator.

D

Data retention

How long the platform stores conversation data after account deletion. Reputable platforms publish a clear window (7-30 days). Skip platforms with vague or missing retention policies.

Diffusion model

Image generation architecture (Stable Diffusion, SDXL, SD3) that builds images by progressively denoising random noise toward the target prompt. Standard in 2026 AI girlfriend image generation.

Discreet billing

Card statement shows a generic descriptor rather than the platform name. Privacy feature for buyers who share bank statements with partners or family.

E

Encrypted chat

Conversation data encrypted in transit (HTTPS) and at rest (server-side encryption). Standard practice on reputable platforms in 2026.

F

FaceLock

SoulGen AI feature that locks a character face across multiple renders. Solves the consistency problem inherent to default diffusion models.

Free tier

Pricing tier at $0. Quality varies from genuine free experience (Kindroid) to limited trial (most others). The /best/free hub covers the 8 platforms with usable free tiers in 2026.

G

Group chat

Conversation with multiple AI personas in one scene. Nomi AI supports up to 10 Nomis simultaneously, the category exclusive in 2026.

I

Image generation

The process of creating images from text prompts. AI girlfriend platforms use diffusion models tuned for character consistency. SoulGen AI FaceLock is the leading consistency tool in 2026.

L

Limitless mode

Janitor AI feature that removes platform content filters after ID verification. Lets writers explore adult roleplay without watered-down responses. Platform-level bans (minors, real-public-figure) remain.

Live Action

Candy AI feature shipped February 2026 for short animated video clips of the AI girlfriend character. Paired with photoreal images and voice messages.

LLM

Large language model, the AI engine that generates text replies. Modern AI girlfriend apps build on GPT-4/5, Claude, DeepSeek, or proprietary models like SpicyXL (141B parameters at SpicyChat AI).

Long-term memory

Equivalent to persistent memory. Stored in a database the AI checks before each reply. The Nomi AI three-tier architecture is the most sophisticated 2026 implementation.

LoRA

Low-rank adaptation, a fine-tuning technique used in image generation. Seduced AI Model Trainer creates private LoRAs ($60-65 one-off) that lock a buyer-trained character across renders.

M

Memory drift

When the AI starts misremembering or fabricating facts about the buyer over time. Caused by long-term memory storage corruption or context window overflow.

N

NSFW

Not safe for work, the standard category label for adult content. AI girlfriend NSFW unlocks behind age verification on most platforms. Universally bans content depicting minors.

O

Opt-in adult content

Platform model where NSFW unlocks via toggle after age verification. Default state is SFW. Standard pattern across mainstream AI girlfriend apps.

P

Persistent memory

Long-term memory that survives across chat sessions. The AI remembers facts you shared weeks ago. Native on Nomi AI, Kindroid, Replika, Lovescape. Most cheaper apps ship session-only memory instead.

Persona

The character configuration on an AI girlfriend platform. Defined by name, traits, backstory, and system prompt. Platforms ship persona builders that abstract the configuration into UI choices.

Photoreal

Image style targeting realistic camera-grade output. Candy AI leads on photoreal quality in 2026; SoulGen AI competes with FaceLock consistency.

Pose Editor

PromptChan feature for posing characters in action scenes while preserving 95% of facial identity. Released in the 2026 update.

R

Reference upload

Buyer-supplied image used as a face anchor for subsequent renders. aiAllure accepts up to 4 reference images per companion. Improves character consistency without full LoRA training.

Roleplay

Conversational play where buyer and AI take character roles in a scene. Central use case for many AI girlfriend platforms. SpicyChat AI and Janitor AI are roleplay-heavy.

S

Selfie request

PromptChan 2026 feature where the AI sends "live" photos during chat without explicit prompt. Adds visual presence to the conversation.

SFW

Safe for work, no adult content. Replika, Anima AI, Character AI are SFW-only platforms in 2026.

Short-term memory

The LLM context window, 8K-32K tokens of recent conversation included in each reply. All platforms ship short-term memory by default.

System prompt

Hidden instruction sent to the LLM with every chat that defines the character. Typically 200-2000 tokens including name, personality, backstory, and behavioral rules. Buyers never see it directly but it shapes every reply.

T

Token

Unit of text the LLM processes. Roughly equals one word. Pricing on token-economy platforms (Candy AI premium, Nomi AI) scales with token usage.

Token economy

Pricing model where buyers pay per token used rather than a flat subscription. Real spend on Candy AI runs $25-60/mo with active token use on top of $12.99 base.

Tone drift

When the character personality shifts mid-session. A stoic warrior becomes peppy by turn 50. PolyBuzz holds tone for ~150 turns; weaker platforms drift sooner.

TTS

Text-to-speech, the engine that converts text to audio. ElevenLabs-tier TTS is the 2026 quality bar.

U

Uncensored

Marketing label for AI girlfriend apps with minimal content restrictions. True uncensored requires bringing your own LLM API (Janitor AI). Or platforms that ship with no filtering by default (Muah AI, Nastia).

V

Video generation

The process of creating short video clips from text prompts. 4-6 second clips at 720p is the 2026 standard. Candy AI Live Action and aiAllure Video Model V5 lead the category.

Video Model V5

aiAllure 2026 video model with 10x faster rendering than V4. Pro plan unlocks unlimited image and video generation.

Voice cloning

Training a voice synthesis model on a buyer-supplied audio sample. Rare on AI girlfriend apps because it raises non-consensual deepfake risks. Most platforms ship preset voices instead.

Voice latency

Round-trip time from your message to AI voice reply. Under 2 seconds feels real-time. Nomi AI hit 1-1.5s in early 2026; Secrets AI runs 0.9s.

Voice synthesis

Text-to-speech rendering of LLM replies. Quality varies from flat narration to genuine emotional inflection. OurDream took the Best Voice Chat 2026 award for emotional range.