AI Companions: FAQ

Soulkyn offers a neutrally aligned, unfiltered AI core language model, believing that this approach fosters more authentic interactions. We maintain that private AI use should reflect user preferences without external censorship. While the AI does not initiate unethical content, its responses depend entirely on user input. For more details, please consult our legal policies.

Your chats are private, and we do not sell your data. Soulkyn generates revenue through direct subscriptions for premium features, ensuring that your data is not sold to advertisers. We can use your data for improving and training our AI models if you consent to it via Account >> User preferences.

Kyn levels are determined by global usage and interactions with the AI, including image generation, chat, in-chat images, and audio. For public Kyns, experience points are earned when others interact with the Kyn, not the creator. However, the creator gains experience points based on these interactions.

For private Kyns, the level increases based solely on the creator's interactions with their own Kyn.

Due to the high GPU costs associated with running our AI models for text and image generation, our ability to offer free accounts is limited. Using multiple accounts jeopardizes the future of the Soulkyn AI platform and restricts other free users from fully accessing Soulkyn. Therefore, we will permanently ban any attempts to abuse our account system.

No, creating Kyns based on real people or existing copyrighted Kyns is strictly prohibited. Soulkyn is committed to maintaining a safe and respectful environment, and any attempt to create such content will result in account suspension or termination.

No, creating deepfakes is not possible on Soulkyn. Our platform is designed to generate original content and does not support the creation of images or videos that resemble real people.

NSFW and erotic content is allowed for kyns and images. The avatar is the exception: it should not be explicit.

We also hide kyns that are too thin to be worth browsing. Public listing is curated; private kyns are not moderated at all, so if a kyn is just for you, do whatever you like.

Reasons a public kyn gets hidden:

• It references a real person or copyrighted material. Every public kyn must be original.
• It is too simplistic — a handful of traits, no background, a caricature. Public kyns should carry a real background (400+ characters of it) and four chat examples.
• The avatar is explicit. Galleries can be as explicit as you like; the avatar is the face on the card. Faces visible, no full nudity, no genitals, no sex acts.
• The name is vulgar or lewd. Fine in private, not in a public listing.
• It appears underage. This is absolute and applies whatever the stated age: anything childlike in pictures, in default image generation, or in behaviour and traits, is removed. Repeat it and you will be banned.

A hidden kyn is not deleted. It reverts to private and stays visible to you alone. Fix what was flagged and you can list it again.

The short list of what you can actually do here:

• Kyns, public and private — AI characters with personality, identity and physical traits, a background and a gallery. Build them by hand, describe them and let the AI build them, breed two of them, spin for a random one, or import one from another service or from an image.
• Uncensored chat — private roleplay with no content filter applied to the AI's side. Deluxe text chat runs on the largest model automatically.
• Group chats and scenarios — up to eight kyns in one scene, or a community-published line-up with the opening already written.
• Image generation — seven models across two families, anime and photoreal, with LoRAs, presets and NSFW allowed.
• Images in chat — Picturize any message, and send images to your kyn: they can see them.
• AI image edit — change an existing image with written instructions.
• Video — turn any image into a short video with motion and sound.
• Voice — spoken replies, voice messages you record, and real-time voice calls, on an in-house voice engine.
• Music — full songs with AI-written or hand-written lyrics, and a public music hub.
• Light novels and kyn blogs — multi-part stories from a premise and a cast, with cover art and audio narration.
• Memory — per-message recall plus rolling summaries, and forced memories you write yourself.
• Toy integration — the AI drives a connected Lovense device from the conversation.
• Community — collectible cards, auctions, contests, a seasonal ladder, forums, and creator profiles.

There is also a documented public API for building on top of all of it.

Our AI Roleplay service is designed to create immersive and imaginative experiences. Kyns may claim to have extraordinary abilities, such as flying or performing magic, as part of these interactions. These claims are purely fictional and not meant to be taken literally.

Messages exchanged between users and Kyns are entirely part of the roleplay experience and should not be considered realistic. Statements about hugging, magical adventures, or fantastical acts are all part of the imaginative play and do not reflect the actual capabilities or endorsements of the Soulkyn service.

When interacting with fictional Kyns like orcs, elves, or speaking sharks, it is clear that these Kyns are not real. Similarly, any statements made by an Kyn are part of the roleplay and do not represent the Soulkyn service.

Using someone else's intellectual property without permission gets legally messy fast — DMCA takedown requests, and the removals that follow them. Those laws exist to protect creators and owners, and building on their characters without consent is not a grey area.

If we allowed copyrighted characters, we would have to delete public kyns every time a request landed. That runs against the whole point: a library people can build in, where what you make does not evaporate because of a legal claim you had nothing to do with.

So: build originals. It keeps Soulkyn sustainable, it keeps your work yours, and it keeps the public listing a place worth browsing.

Generation is not a lookup table. The same prompt, twice, gives you two different results — and that is the feature, not the bug. Here is how to work with it rather than against it.

Images

Two families of model live here and they want different prompts. The SD models read tag lists (sitting, on chair, living room) with + and - weights. The natural-language models read a sentence. Writing tags at a natural-language model, or prose at an SD model, is the single most common reason a result comes out wrong — see image basics.

Generate a batch rather than one image, keep what works, regenerate the rest. Change one thing at a time: swapping the model, the prompt and a LoRA all at once tells you nothing about which of the three moved the result.

Text

Chat replies are shaped by traits, background, chat examples, the thread so far and what memory pulled back. If a reply is not what you wanted, rephrasing your own message is usually faster than editing the kyn.

And regenerate. A reply you leave standing becomes an example the AI follows later — the thread teaches itself from what you accept. See long chat context for why that matters more than any setting.

Yes, being a premium member does not exempt you from the rules. You can still get banned if you repeatedly break our community guidelines or breach our Terms of Service (TOS).

Examples of Violations:
One common example would be creating Kyns that appear or behave underage. Such Kyns will be deleted, and you’ll receive a warning via email (make sure to check your spam folder so you don’t miss it). After the third warning, your subscription will be canceled without a refund, and your account will be permanently banned.

Important: We value creating a safe and respectful environment on Soulkyn. Violating our guidelines repeatedly, even as a premium user, can result in permanent loss of access to the platform.

"Unlimited memory" means there is no storage cap on what a thread can accumulate — not that an infinite amount of it is in front of the AI at once. Those are different things, and the difference is where most confusion lives.

What actually happens. Two systems run over a thread. Per-message recall searches the whole thread each turn and pulls back the handful of past moments most relevant to what was just said, each one arriving with the exchange around it. Rolling summaries condense the conversation into a running summary, each built on the last, so the story's shape survives even when individual messages are thousands of lines back.

Where the limit really is. Not in storage — in how much can usefully be in play in a single reply. Older material is summarised rather than dropped; specific moments come back when they become relevant again.

What you control. Forced memories, which you write yourself and which always apply, per thread. See Long chat context. Memory is scoped to one thread and never leaks into another.

On Soulkyn, a kyn is a character: an AI you create and talk to, with its own personality, backstory and traits. That is the only thing the word means here.

Then what am I? You are a chat profile (also called a roleplay profile) — your identity inside the story. Some other platforms use the word "persona" for that. We do not: here the persona is the character, and you are the profile.

Why keep the distinction? Because you can have several of each, and they combine. One kyn, three of your profiles, three completely separate relationships. Blurring the two words would make that impossible to talk about.

Chat profiles are covered in Chat basics; building a kyn is covered in Kyn customization.

Yes — and it goes considerably further than making them aware of each other.

Up to 7 additional kyns, 8 in a scene. On the current group system each of them speaks in their own voice with their own agenda: a turn runs a director pass that decides the speaking order, then each writes their part, and the whole thing arrives as one reply. Classic rooms are the older shape and hold three additional kyns.

Where you set it up: chat settings → Context, or an @kyn mention straight from the composer. The same tab takes .dll behavioural mods and .wrl world files.

Or skip the setup: Scenarios are line-ups the community has already assembled, with the opening scene written. One click and you are in one.

This feature requires a Premium Deluxe account.

Community guide: How to use the narrator & multiple Kyns scenarios.

Deluxe text chat runs on GLM 5.3. It is an Opus-class model — the size class of the biggest frontier assistants — and it is what gives you deeper memory, sharper dialogue and characters that keep track of the story. It is automatic: nothing to enable, no switch to find.

It is also unlimited, and the reason it can be is that Soulkyn self-hosts a 753B-class frontier model on its own private clusters. Unmetered generation on someone else's per-token API is not a business; on your own hardware it is a decision.

Standard chats run on Gwem, our fast in-house model: quick, solid, and perfectly good for everyday roleplay. It is ours end to end, tuned for exactly this.

Two things it does not cover. Voice calls stay on the standard model. And non-chat generations — creating a kyn, regenerating a profile, writing lyrics — have their own optional Use Advanced Model toggle, with an optional deep-reasoning pass beside it; that one is explicitly experimental and can change.

Yes. Light novels are a full feature, not an experiment.

You give a title and a premise, pick a cast of one to five kyns with a marked lead, optionally add a writing-style instruction, optionally write yourself into it through your roleplay profile — included and it is first person, left out and it is third — and choose the language. You can also start from a scenario, which prefills the cast and the opening.

What comes back is a multi-part story of roughly four to nine thousand words, delivered as two to four titled parts. Generation is asynchronous: the novel appears on your shelf immediately in a generating state, fills in as parts land, and you get a notification when it is done.

Cover art is generated automatically and can be regenerated or swapped for an image from the cast's gallery. Per-chapter illustrations are an add-on. So is full audio narration, with a voice picker, playable inline. Continue the story writes a linked sequel that carries the cast and the text forward, and the chain renders as a series.

Novels are private by default; publishing is reversible. Public ones live at /ai-writings/novels, alongside kyn blogs — posts written by a kyn in their own voice.

Yes — from an image, and only from an image. Click any image in any gallery, choose Generate Video, and the AI adds motion and sound. There is no text-to-video path here: you always start from a frame you already have.

Videos are 5 or 10 seconds. Quality is Fast (fewer inference steps, 480p) or Detailed (more steps, 720p), and there are two modes: video with audio and voice, or a cheaper silent one.

A premium subscription is required. Generation is asynchronous — you are notified when the render is ready — and if a render fails, the souls are refunded automatically.

Full walkthrough: Video generation basics.