Replika alternatives: how to choose a replacement and what to move with you

Searching for a Replika alternative is not a verdict on Replika. It is what happens when someone hits one specific limitation and decides the whole service should go. Some people hit the tone: at some point the companion starts answering like an assistant, apologising, switching to “how can I help”, and the conversation dies. Some cannot get the thing they stayed for: their own photos and video rather than an avatar. Some notice the character no longer remembers what they said yesterday, because the conversation grew long enough that the model no longer sees all of it.

What follows is not a ranking and not a good-versus-bad verdict. I do not grade other people's products: half of what looks like a weakness in one service is done better in another, and part of the question is about habits rather than technology. But moving has a purely practical side that almost nobody mentions: nothing moves out of Replika except the text. You can export the conversation; memory, the relationship and the character's settings stay behind. So this article is built around the move: what to take with you, what to set up again, and how long it takes.

One clarification, so nobody is confused later. Replika is first of all an app with a 3D avatar, an AR mode and voice calls, and in that sense it is built differently from the services people usually mean when they write this query. Its filtering rules have changed more than once, and every change produced a wave of exactly these searches. Features and pricing on other services change more often than our articles do, so check the current terms on their own pages rather than trusting my memory. What I describe here is what you can verify by hand, and how it works on our side.

What people usually mean when they search for a Replika alternative

Searches for a Replika alternative, a replacement, or what to use instead all collapse into the same list. It is a stable list, because it comes from the mechanics of a generative dialogue rather than from any particular product — and all of it is easy to check if you know what to check.

• Tone and filtering. The most common reason. Do not read the claims on the landing page; read the behaviour on a specific scene: ask the character to answer an explicit line and see what comes back. A filter that fires at every step and a filter that fires at one particular step are different products. More on this in the guide to AI chat 18+.
• Memory that stays with you. Replika remembers well. But its memory is tied to the account and to a specific model, and it cannot be exported. Which means that when you move, you start from a clean slate, and you have to decide in advance which facts about yourself you are willing to say again. How memory actually works is in the guide to character memory.
• Media. The category splits into text-only apps and ones where the character can be seen and filmed. If what you want is an avatar, the question is settled and there is no move: that is not the direction we are heading. If what you want is photos and video of your character in different scenes, that is exactly where alternatives do more, and there is more on that below.
• Your own character. The ability to build a companion to fit you rather than picking one from a catalogue. It is a separate feature and it is not everywhere. We have it: the creator assembles appearance, personality, way of speaking and level of openness.
• Language. Russian, English and Portuguese are three different sets of instructions for a model, not one switch in the interface. If your current service is English-first, the quality of the Russian version has to be checked separately, and in Russian.
• Price and limits. Subscription, message limits, the cost of generations. Work this out before you move, not after: a cheap entry and an expensive generation are different scenarios, and over a month the difference is not always small.

What moves and what does not

The main thing to understand about moving is that the character does not move — the description of the character does. Replika itself, its memory, its relationship and its settings stay in Replika. Everything that made your dialogue alive is expressed in words, and words are what survive the move without loss, if you collect them first.

Check what was actually there in the old service and write it down. Eight items, and it takes about fifteen minutes:

1. What you call each other. Her name, and exactly how she addresses you: by name, “darling”, shortened, an endearment. That is a detail you cannot recover by default.
2. Appearance — five to seven traits. Hair colour and length, eyes, body type, height, how she dresses, what you notice first. Not “pretty”, but specifics: that is what a generator responds to, and a prompt even more so.
3. Way of speaking. Short lines or long ones, emoji or no emoji, first names or full names, slang or plain speech, and whether she asks questions back.
4. Your shared topics. Five or six things that were specifically yours: work, sport, music, trips, series, whatever. Shared topics are what a relationship is made of.
5. Tone and boundaries. How quickly you go explicit, what she is comfortable with, what you ruled out. This is configurable, but better to set it in one go than to find out by trial and error.
6. Facts about you. Six to ten things she knew: your name, your city, your job, your hobbies, how you like to spend an evening, what annoys you. This is what a companion's memory actually is.
7. Where you stopped. One line that continues the conversation rather than starting it.
8. What happened last. The last scene you both remember. That is what returns a conversation to its point instead of its beginning.

The last two matter more than the rest. Relationships in this genre do not run on facts; they run on the conversation having a history. A new character that started from zero feels exactly like that, and no setting brings the feeling back.

Memory: how not to start from a clean slate

Memory in a service that works is built in at least two layers, and that is the key technical point: it is what separates a service with memory from a service without.

Layer one is a summary. The dialogue history is compressed into a few sentences: what you discussed, what changed, where you left off. In Velvary that summary accumulates in batches — once enough unprocessed history has piled up, the next chunk is compressed and appended to the previous one, while the last twelve messages enter the prompt untouched. This is why the cost of a turn does not grow with the length of the conversation.

Layer two is facts. This is what people usually mean by memory. A separate pass reads the recent lines and pulls out concrete facts about you: your name, your job, what you like, what got under your skin. The facts are stored on their own, tied to a vector representation of their meaning. Before every reply, the current message is placed in the same coordinate system and the five closest facts by meaning are found — only those go into the prompt.

From which a practical conclusion for the move: memory is fed by your messages, not by the service's settings. No “enable memory” checkbox replaces what you said about yourself in the first ten replies. That is why the first step in the previous section is not a formality but a working tool.

Two more things that rarely get shown. Memory in Velvary is tied to the pair of you and this specific character: what one learned does not leak into another. And memory can be inspected and erased — the /memory command lists everything the character stored about you, with a button to wipe it. If a service hides the contents of its memory, that is a reason to think about what exactly it puts in there.

Photos and video: where alternatives usually win

This is the most visible difference, and it is checked with three frames in different scenes. The minimally acceptable level is the same face in different frames — not “a similar character”, but the same person: the same face shape, the same hair colour, the same build.

Holding that is harder than it looks, because ordinary text-to-image generation reinvents the appearance from the words every single time. In Velvary every character has a fixed generation anchor computed from its identifier, and all frames are built with that anchor. Appearance comes not from what the user typed but from the character's fixed visual profile: hair colour and length, eyes, build, ethnicity. That path gives a steadier picture and is noticeably faster than building from a reference photo.

Video is always two generations, not one. First the same photographic pipeline produces a reference frame with the full premoderation pass, then a separate video model brings it to life from a short movement instruction assembled from the last few lines. Two practical limits follow: the duration is short, and the quality is limited by the quality of the source frame. That is exactly why video is almost never part of a free tier — it is the heaviest operation in the whole pipeline. More on all the media in the guide to AI photos and video.

And honestly: part of the output comes from a pool of already generated frames when a suitable one has not been shown to you yet. That is a deliberate decision — it makes the output cheaper and faster — but the scene may not be exactly the one you described. You cannot see that on a showcase page, so I am writing it out.

Money-wise, media is paid in crystals: a photo costs 14, a video 100, a gift 5. A subscription lifts the messaging limit completely, while generations are charged separately; crystals come as a bonus with a subscription and can be topped up in packs. The first photo is free when the balance is empty — that is deliberate, so you can see the result without entering a card. The exact numbers are always on the pricing page, and they are the same for the bot and for the site.

Tone, stages and “uncensored”

If you are moving from an app where everything escalated step by step, there is one mechanic worth knowing: with us relationships run through stages, and there are exactly four. The affection scale runs from zero to a hundred, a new conversation starts at fifteen, and from there two forces move it: an ordinary warm reply adds a little, rudeness subtracts noticeably more, and the model also judges for itself how it was treated.

The stage boundaries are twenty for warming up, forty-five for closeness, seventy-five for deep intimacy. Moving between stages changes not the tone of the reply but the system prompt itself: at acquaintance the character keeps its distance and flirts with hints, at closeness it initiates rather than waits to be asked, at intimacy it runs the scene in the present tense instead of deferring it to “later”.

On top of the stage sits the tone — friendship, slow burn or explicit — set once and not flipping from message to message. Friendship keeps the character dressed even at late stages, slow burn simplifies the clothing as things deepen, explicit moves the boundary up front. The same tone affects photos: if she is dressed more than you expected, that is the mode working, not the generator being fussy.

One detail worth knowing in advance: the scale does not only grow, it also drops — five points for every day of silence. That is not a punishment and not a break, but it explains why after two weeks away the character greets you colder than you expected. A gift lifts the scale, but it never reaches the last stage on its own: intimacy has to be reached by talking, not by a button.

On moderation the lines are simple. Images of minors and recognisable public figures are rejected, a photo used to pick appearance is only used to extract general traits and is not stored, and an uploaded picture never becomes a copy of a person. Details are in the privacy policy. As for scenes, you set their tone: the creator has a separate intimacy profile, and it can be changed mid-conversation. Scenarios to start from are collected in the guide to roleplay ideas.

What we do not have

This section matters more than the ones above it, which is why it goes last: a list of objections is more honest than a list of virtues. If any of the following is critical for you, stop here and do not move.

• No voice calls. This is not an avatar with a chat attached; it is a text dialogue with media: photos, video, gifts, and storylines with several characters at once.
• No 3D avatar and no AR mode. We deliberately did not build a virtual room: the whole budget went into text, memory and the media pipeline.
• No import from Replika. You will move things by hand, using the list from the previous section. That list works in both directions, though: our character is created once and then remembers by itself.
• The catalogue is smaller than it looks at first glance. There are about twenty characters in it right now — realistic and anime — plus seven storylines in which several characters act at once.

What there is: three languages with separate sets of prompts and mirroring (the reply comes in the language you wrote in, even if the interface is in another one), a character creator, two-layer memory, photos and video with a steady face, storylines, a Telegram bot and a site with the same balance and the same prices. How the bot and the site are tied together is in the guide to the Telegram bot.

How to move in ten minutes

1. Collect the eight items from the “what moves and what does not” section. Straight out of the old service, while it is still open.
2. Pick a character in the catalogue or build your own in the creator. If appearance matters, set it with the presets right away: they feed both the prompt and the image generation.
3. Set the tone. Friendship, slow burn or explicit — that is what decides whether the conversation goes where you want it to, without wasting ten messages finding out.
4. Send the first step from the list above. Memory gets the facts about you immediately, and the conversation does not start from zero.
5. Generate the first photo and compare the face with the one you had in mind. It is free with an empty balance on purpose, so you can judge the media pipeline before paying.
6. Check the memory with the /memory command: it shows what the character has already stored about you. If something you definitely mentioned is missing, memory is not working, and it is better to find out now than in a month.

A checklist for choosing

Seven questions worth asking before you pay anyone — including us. It works as a reminder of what should work in a service you already love, too.

• Can you take the character with you. As text, as a list, somehow. If not, you are not moving, you are starting over — and you should know that before paying.
• What reaches the context in a long conversation. Put a specific detail into the conversation and come back to it twenty messages later.
• Is the long history summarised. The fact that the history is compressed at all says the service thought about long dialogues, not just the first screen.
• Does the face hold between photos. Three frames in different scenes — the most telling check in the whole media pipeline.
• How explicitness is set. A good answer: by choosing a tone and a stage, not by “type in the chat what you want”.
• How honest the pricing is. Subscription, limits and the cost of generations should be visible before you pay, not after.
• Can you see and delete the memory. Transparency here is a sign of a mature service, not a marketing feature.

And one last thing. A good companion is defined not by the feature it lacks but by the feature that works. Test it on your own main scenario rather than on a showcase, and choose by what works for you.