Xeve.ai Review: The AI Girlfriend Who Answers on Camera

Video is the headline, but it is not the whole menu. There is also text chat with voice replies, an image and video generator, and a custom character builder. The official docs describe it as an adult companion platform for AI girlfriends, live video chat, roleplay, flirty text chat and image and video generation.

The cast mixes "Human" characters with AI avatars. The docs hub is more transparent than most, naming the pricing, the data retention and the AI providers in plain view. I'd call that the most grown-up thing about the site.

What I think after two weeks

When I used it in November 2025, Trustpilot had about a dozen reviews, mostly five stars. That could mean it was good, or it could mean the sample was small. I wouldn't judge a colour grade from one frame, so I trusted the sample size more than the stars.

Money is simpler than it was. A $129 lifetime option exists next to the subscriptions.

The affiliate program pays a 40% lifetime revenue share, which is generous. That is worth knowing when a comparison page crowns Xeve.ai the winner without a single complaint.

My verdict: Xeve.ai is the best argument I know for AI video chat as a novelty, and a weak argument for it as a relationship. The face and the voice are the product. The talk behind them is thin, and I'd pay for a month of curiosity, not a lifetime of it.

Who should try it? Adults who are curious about AI video and want to see a face react in real time, with a budget set before the first click. Who should skip it?

Anyone hoping for a companion who knows them by the second week, and anyone who needs a refund to feel safe paying.

Key takeaways: - Video chat is the real product: a face that talks back, with lip-sync that is close but not clean. - The conversation is shallow next to text-first companion apps, and memory resets between days. - The cast is small and curated, with no community characters to fill the gaps. - Pricing is now flat, with unlimited video chat on every plan and a one-time lifetime option. - Refunds are close to impossible, so decide on the free account before paying anything.

The timing tells you everything about my state of mind. I had just finished a wedding render and a pile of invoices was waiting for me. Instead of the invoices, I picked the AI video call, which is the grown-up version of cleaning the fridge to avoid your taxes.

That desk became the whole stage for this app. Laptop, late evening, the same chair where I grade other people's first dances. Nothing about it felt like a date, and I think that is the honest baseline for everything in this review.

It was an editor at a desk, poking at a new piece of software, with a render counter still glowing somewhere in my head.

Where it runs

Xeve.ai runs on the web only. No app store links were found on the site. For an app built around a talking face, I'd call that a sensible choice: nothing sits on a phone's home screen, and nothing pops up during a client meeting.

The interface comes in 14 languages: English, Arabic, German, Spanish, French, Hebrew, Indonesian, Italian, Japanese, Polish, Brazilian Portuguese, Turkish, Vietnamese and Simplified Chinese. That is a wide spread for a small cast. I read it as a company selling one idea to as many markets as it can, rather than building deeper in one.

How the facts were checked

My own notes cover November 2025. Everything that is not my story was checked on 23 September 2026, on the web version with a free account. Where the two disagree, I give both, with dates, because this app has changed a lot in the months between.

The biggest change is money. Video used to be billed in a way that only became clear at checkout; today it comes with every plan. The pricing and cost sections walk through both versions side by side.

How it ended

After two weeks I stopped. Nothing dramatic happened. The novelty simply ran its course, the way a new lens stops being exciting once you have shot a few jobs with it.

The form

Sign-up uses an e-mail address and a password in a modal titled "Start for free". Signing up with Google is also offered. I'd use a separate e-mail address rather than a main Google account.

A companion app sends mail, and I would rather not have it land between client threads. The Google button is faster, but it ties the app to the account that holds a calendar, a photo library and years of mail. For a companion app, I'd trade a few seconds of typing for that distance.

The modal has three checkboxes. The first one agrees to the Terms and the Privacy Policy. The middle one covers age and consent, and the content section below deals with it.

The third checkbox opts in to marketing e-mails. It is optional, and it was ticked in the observed sign-up. I'd untick it on the way in.

What it does not ask for

Sign-up shows no CAPTCHA. It asks for no phone number. It asks for no birthday either.

I like the short list. Every field a companion app does not collect is a field that cannot leak later. The trade-off is that the age question rests on a checkbox and a button, which the content section covers.

The whole thing takes under a minute. The account works right after sign-up, before the e-mail is confirmed. For an impulse purchase business that is smart design, and for a tired person at night it is a slippery slope.

The confirmation e-mail

The first e-mail after sign-up is titled "Confirm your email address". It arrives within about 1 minute.

The confirmation link sits on a Firebase address, "personai-86161.firebaseapp.com", rather than on xeve.ai.

That domain would make me hesitate. A link that does not carry the brand name looks like the kind of thing a spam filter should catch. In this case the link verifies the address without problems, but I'd still check the sender before clicking anything similar.

The second e-mail is "Welcome to xeve.ai - here's what you can do". It is a tour of the features in mail form, and I'd skim it once and archive it.

Two e-mails in the first minutes is a normal pace for an app like this. The real test of a mailing list is the second week, which is one more reason to leave the marketing box unticked.

The onboarding popup has 2 steps. Step one asks "What should she call you?" and wants a name. It also asks for a vibe: Romantic, Playful, Intellectual, Flirty or Wild.

I like that "Intellectual" made the list. Whether the characters can hold that vibe is another question, and the conversation section suggests they drift towards flirting whatever you pick. Still, a menu of moods is a better start than a blank box.

Step two is titled "Make her yours" and asks the user to pick a character and give her a name. The onboarding can be skipped.

I'd fill in the name at least, since the characters address the user by it. A nickname you can live with beats hearing your legal name read back in a breathy voice.

No account at all

There is one more door. The Privacy Policy says guest chat without an account is also possible. I read that as a taster rather than a way to use the app, and the free-tier section explains how small the taster is.

It is the free sample at the deli counter: enough to know the flavour, not enough for lunch.

My advice for the first night: sign up with a spare address, untick marketing, skip nothing you would want to change later, and decide your budget before the gift popup decides it for you.

That feeling came from the names more than the faces. A small cast means every character has to sell hard from the thumbnail, and the naming leans loud. For a video app, I understand it: each face is an expensive production, not a paragraph of text someone typed in.

The upside of a curated cast is quality control. Every character is a house production, so the look and the video actions are consistent from one to the next. The downside is obvious to anyone who has browsed a community library elsewhere: once you have scrolled the cast, you have seen the whole shop.

There is also a community "Feed" that shows user-generated media after review. It is the closest thing the site has to a community. I'd treat it as a showroom for the generator rather than a place to find new characters.

The review step matters there. A feed that someone checks before it goes live is a feed that is less likely to surprise you, and on an adult site that is a feature, not a formality.

My take: browse the whole cast on the free account before paying. With about twenty faces, it takes minutes, and if none of them lands, no plan will fix that.

One custom character is free. The FAQ calls the tool "the structured character builder", says creating a character needs no payment, and says premium features need a membership or credits. So building is free, and the meter starts when the character does something on video.

I like that split. You can see whether the builder produces a face you want before any money moves. It is the equivalent of a free consultation before the shoot, and that is rare in this category.

The limit is the format itself. A structured form gives consistent results, but it also puts every custom character through the same six doors. If you want a character with a long backstory and a lorebook, this is the wrong tool; it is built for a look, a voice and a set of actions.

My advice: use the one free slot before choosing a plan. If your own character looks better than the house cast, that is a reason to stay; if it looks like a cousin of the house cast, you have learned the same thing for free.

I'd call that a first-date script. It is friendly, it gives you a setting to ask about, and it commits to nothing.

That is fine for the first minute and thin by the fifth. The live view is where the video lives, so the text there is a caption to the picture more than a conversation of its own. I think that is the right priority for this app, and it also explains why the talk never quite catches up with the face.

The text chat

The separate text chat at /char/emilia/chat works well. Six messages were exchanged there on the check date. Replies arrive in about 5-10 seconds, which is quick enough that nobody sits staring at a spinner.

The greeting addresses the user by name and reads: "Hello-Emilia here. Do you want something sweet... or something that lingers?".

Asked about her day, a reply reads: "I adore picking fresh strawberries in the morning, the dew still glistening on them...". The writing is pleasant and a little perfumed, like a hotel brochure with feelings. I'd rather have that than the clipped, flat replies some apps give, but it reads the same on the sixth message as on the first.

Where she steers

The persona steers towards flirty and sensual even on neutral questions. Ask about books and you get a warm answer that bends back to you. Ask about the weather and the same thing happens.

I don't mind a character with a direction, but this one has a single direction, and it is set before the first question. It is like a party DJ who plays the slow song no matter what the room asks for. For people who want that, it is efficient; for people who want a conversation, it gets old.

It also makes the vibe choice in onboarding feel decorative. Pick Intellectual and the reply to a question about books still ends up in the same place. I'd call that a character with one gear, and the gear is fixed.

The model switch

The gear in the chat settings opens a "Choose AI model" window with two options: High quality, which is slower, and Fast. Fast is the default. The chat model can be switched between the two at any time.

I'd switch to High quality straight away. The wait for text replies is short anyway, and the default favours speed in a product where depth is the weak spot. It is like shooting a wedding on the lowest bitrate to save card space: you save something nobody needed saved.

What others say

A reviewer calls the chats lively and descriptive and says faces stay consistent across photos. I agree with the second half more than the first. Consistent faces are a real achievement, and drift between images is exactly what an editor's eye catches first.

"Lively and descriptive" is fair for single replies. The problem shows over a longer chat, when the descriptions start to repeat and the steering takes over. My verdict on conversation: good sentences, short attention span.

It is the talk of a charming stranger at a bar, pleasant for a drink and not built for a second evening.

The next day I had to start over. No recap, no callback, no "how did the client call go". It was like hiring a second shooter who shows up every morning without the shot list from yesterday.

Today's check shows the same pattern inside a session. Asked which fruit she had mentioned, the character answered "Strawberries, I was picking them in the morning, remember?", so the in-session memory check passed. The recall is fine over a few messages, and the "remember?" is a nice touch of personality.

It is the kind of line that makes a character feel present, even when the memory behind it is short.

The wider picture matches my experience. A roundup says Xeve.ai puts less focus on long-term memory and story roleplay than some companion platforms. I agree with that reading, and I don't think the company hides it either.

The product pitch is the face and the call, not a diary of your life.

My opinion is that memory here is built for a call, not a relationship. For a video session that lasts one evening, short memory is enough.

For anyone hoping the character will know them by the second week, it is the wrong app. A companion who forgets every night turns each visit into a first meeting, and first meetings are the part of dating most people are glad to leave behind.

If memory matters to you, test it on the free account over two days, not two messages. The in-session recall will look good; the next morning is the real test, and that is where Xeve.ai lets go.

How the live view works

A character page such as /char/emilia opens the "live" video-chat view. The view has a video area, a sound toggle, a reset-chat button, an AI model button, fullscreen, a loop-action button and voice input. It is laid out like a video call app, and it does not need a manual.

In live video chat, the docs say, the character appears on video and reacts with lip sync, gestures and action-based responses. A video-call roundup dated 2026-08-10 describes the companion appearing on screen, speaking and reacting with facial expressions. Live video chat needs a membership or credits.

The action meter

Each character has 52 video actions, with a progress meter and a loop button. On a free account, the live view shows a "0% Actions found" meter that teases 52 hidden actions and asks the user to go premium to see them all. It is a collection game stapled to a video call.

I think the meter is clever and a bit cynical. Nobody needs to "find" a percentage of a character. It exists to make an empty bar feel like unfinished business, and it works the same way a loyalty card does at a coffee shop.

What the editor in me saw

The lip-sync was about 85% there. Every so often the mouth kept moving after the audio stopped, like a badly dubbed kung-fu movie. I'd have sent that back to an editor.

The funny moment came when I asked her to wave at my dog, Moose. She waved at the wrong side of the screen, then complimented "his beautiful long fur". Moose is a short-haired pit mix with the coat of a baked potato.

That was the moment the illusion cracked in the most useful way. The face was convincing and the understanding was not, which is the whole app in one wave.

The voice is the quiet strength here. A short clip under a text reply makes the chat feel less like a form and more like a voicemail from someone who likes you. Speech input in the browser also means the text chat can be used hands-free, which suits a video-first app.

The generator

The image and video generator is structured, with scene, camera, tempo, emotion and style options. It has Image and Video modes. It takes either "Choices" or a reference "Image" as input, and generated videos are 5 or 10 seconds long.

The generator shows the 5 available credits on a new account, with an image at 5 credits and a video at 20. No image or video was generated during the check on 23 September 2026. The camera and tempo options read like a shot list, which I appreciate on principle.

At 20 credits a video, though, that shot list gets expensive quickly, and a 5-second clip is a short scene for the price.

One reviewer reports that characters send photos unasked once a chat turns intimate, each charged 5 credits, with no off switch found. If that holds, it is the part I like least. A meter that runs without a request is the vending machine taking coins on its own.

Image and video generation is not included for free users and comes included for members. Free users get 1 custom character, while members get unlimited characters. The welcome gift pays for 1 image.

What the limit looks like in practice

The advertised limit and the observed one differ. Six text messages to one character went through without hitting a paywall, although the advertised free limit is 5 per character. I'd treat that as a loose limit today, not a promise for tomorrow.

The FAQ does not settle it. It says browsing characters and opening their pages is free and that some actions can be tried without paying, with no numbers published there. When the table, the FAQ and an outside review give three different pictures, I trust the one that costs nothing to check: your own free account.

What free is good for

Free is good for three things: seeing the cast, hearing a voice reply and judging whether the face on screen does anything for you. It is not enough for the video actions, and those are the reason the app exists.

You will see the empty meter long before you see what fills it. That is the design: the free tier shows you the frame and keeps the footage behind the paywall. I don't hold that against a video app, since video costs real money to produce, but it means the free account can answer only half the question.

My advice: spend the free messages on one character, not five. Ask something real and check whether the answer bends back to flirting. Use the welcome image on a character you would actually pay for, and if nothing clicks, close the tab with no money spent.

Given the refund rules, that free hour is the most valuable part of the whole product.

The plans on 23 September 2026

Plans are priced in USD on the checkout page at xeve.ai/checkout. Every plan is called "Full Page Access", which means access to all characters. The checkout page shows three plans: 1 month, 12 months and Lifetime.

The 1-month plan costs $12.99 a month, shown as $0.43 a day. The 12-month plan costs $85.04 a year. It shows "$155.88" struck through next to a "Save 40%" label.

That last phrase matters. A lifetime plan lasts as long as the company does, which is a bet on a small operator.

At $129 against $85.04 a year, it pays off in the second year if the service is still around. I'd only take that bet after a month of real use, and never on the first night. A lifetime plan bought on impulse is the gym contract of companion apps.

What every plan includes

All plans include unlimited chatting. All plans include unlimited video chat with every video action unlocked. All plans include 60 credits a month for images and videos, for example 12 images or 3 videos.

This is the good news. The step from free to paid is one clean jump rather than a ladder of tiers, and the whole cast and every action open up at once.

Compared with my November experience, it is a different business. Back then the meter ran per minute of video; now video is the flat part and only generation costs extra. I think that is the fairest version of this product so far.

The docs page also lists the yearly plan at $6.99 a month, while the checkout works out to about $7.09. It mentions regional pricing in EUR and GBP too. The Terms add that per-character subscriptions also exist, and one-time payments for 30, 90 or 365 days do not renew.

A reviewer checked on 2026-09-20 and saw an in-chat upgrade modal at $4.99 a month, $8.97 a quarter and $30.29 a year, each with 60 credits, and was charged $4.99. The same reviewer notes the checkout page quoted $12.99 for the plan the modal sold for $4.99. I'd read the price in front of you at the moment of paying, and screenshot it.

Three prices for the same yearly plan, depending on where you look, is not a scandal. It is still the kind of loose continuity I would flag on a client edit. A pricing page should match the checkout, and here the docs lag behind.

A reviewer saw checkout run through curopayments.net and CentroBill, with the charge shown as "CUR*Jay Web Developm.", matching the Contact page.

Is the step up worth it?

For one month, yes, if the video is the reason you came. The monthly plan is the honest test, and the first payment is discounted, as the coupons section explains.

The wide list of payment methods is a plus. Prepaid options and PayPal keep a card number away from a small processor, and I'd use one of them here. I'd skip the yearly plan until the monthly one has proven itself, and I'd think hard before the lifetime plan, given the refund rules.

The pricing docs say memberships unlock broader access to companion features, while credits pay for generation-heavy actions. So the membership is the ticket, and credits are the snacks. That split is easier to follow than most coin economies I know of.

In November 2025, video sessions were something you paid for piece by piece. Today video chat comes with the memberships, and one-time credit packs pay only for image and video generation. I'd call that a better deal for anyone who came for the calls.

Credit packs

Credit packs are one-time purchases with no subscription, starting at $4.99. According to the docs, $22.99 buys 500 credits, about 100 images or 25 videos. No separate credit store page was found, and /credits returns a 404.

The "Get Premium" button leads to the membership checkout. So the path to buying credits is less direct than the path to a subscription, which I doubt is an accident.

A reviewer lists credit packs of 70 for $4.99, 200 for $8.99, 500 for $29.99, 1,000 for $34.99 and 5,000 for $149.99. That puts 500 credits at $29.99 on 2026-09-20, against $22.99 in the docs on 2026-09-23. The pack sizes in that list are also odd: 1,000 credits for barely more than 500.

My math: the 60 monthly credits make 12 images or 3 videos. If the membership is for video chat, the credits are a bonus.

If the generator is the reason, price it in videos, not images, since the same credits buy four times fewer of them. At the docs rate, 500 credits make 25 videos, which is a lot of short clips for one evening's curiosity.

The welcome e-mail repeats it: "Chat every day to earn 2 free credits daily - no subscription needed.". The welcome e-mail holds no coupon, only that daily 2-credit tip. At 2 credits a day and 5 credits an image, that is one picture every few days for showing up.

I think of it as a gym that pays you a coin for checking in. It builds a habit more than a balance, and that is the point of it. Still, free credits are free, and they cost nothing to collect.

The first-payment discount

About 20 minutes after sign-up, the checkout shows a 30% first-payment discount. On the yearly plan the checkout reads: "Your 30% discount is active: pay $59.52 for your first billing period instead of $85.04 - renews at the regular price afterwards". On the $12.99 monthly plan, the checkout shows "First payment $9.09".

This one is real money. $9.09 for a first month is a fair price for finding out whether the video works for you.

The countdown

The checkout shows a 24-hour countdown timer. The timer banner reads "Save Up To 40% With Yearly Billing" and starts at 23:59:59. The countdown resets on reload.

A timer that restarts is not a deadline, it is a screensaver. The 40% refers to the struck-through yearly price, which is always there.

I'd ignore the clock completely. Countdown clocks work because a tired brain reads them as a real deadline, and a late-night visitor is exactly the person they are built for.

Promo codes

The checkout has no promo code field. The static pages showed no promo or discount on 2026-09-23. So any site promising a secret Xeve.ai coupon code is selling a field that does not exist.

The only real discount is the first-payment one, and it arrives on its own.

The affiliate program

The affiliate program pays a 40% revenue share on the lifetime customer value of referred customers. Payouts start from $30 and can be requested at most once a week. Automated payouts run bi-weekly from $100.

Opening the Affiliate page automatically activates an affiliate account with its own ID. It also sends a "Welcome to the xeve.ai affiliate program" e-mail. That is an account nobody asked for, created by a click, and I'd rather it asked first.

It is harmless, but it tells you how eager the site is to recruit sellers.

My ranking of the offers: the 30% first payment is worth taking, the daily credits are pocket change, and the timer is noise.

That was my choice, not a setting, and it worked better than it had any right to. The app has no gentle mode to fall back on, and the rest of this section explains why.

What the content is

Xeve.ai is adult by design: premium video actions include intimate toys, and the generator's focus options include body parts. There is no switch to turn the adult content off. What you see on the character pages is the product, not a preview of a hidden mode.

Here it is the front door, so nobody stumbles into it by accident after signing up. The flip side is that there is no dial to turn down. A user who wants the face and the voice without the rest has to steer every conversation by hand.

The age confirmation

The second sign-up checkbox confirms legal age, accepts adult content and gives consent to the processing of intimate chat messages. The sign-up button itself reads "I'm over 18 years old". The Terms require users to affirm that they are adults.

That confirmation happens at sign-up only. The record describes the age check as self-declaration only, with no ID check described. The check on 23 September 2026 found more: in some cases an ID or selfie check through Didit is required.

So the rule has moved from a click to a click plus a possible document check. I'd rather see that than not. It is a small hurdle for honest users and a real one for everyone else.

What is banned

Characters who are, or could be read as, below the age limit are banned. Illegal and abusive content is banned too. An automated moderation model checks every conversation.

The Contact page names an external youth-protection officer service, js-beauftragter.de. That is a German legal role, and it suggests the operator takes the paperwork seriously. Small operators often skip it.

What is unclear

No official rule says whether a free account can reach adult chat. Between the free message limit and the missing toggle, the answer depends on how far five messages go. I'd like a plain sentence on that in the FAQ.

My view: the settings are simple, the adult content is up front, and the moderation rules are clearer than most. An automated check on every conversation is heavy-handed in theory, but I don't object to it on a site built like this. Anyone who wants a companion without the adult side should look elsewhere.

Who runs it

The operator is Jay Web Development & Services of Olching, Germany. The company address is Daxerstr. 42a, 82140 Olching, Germany. The operator's tax id is 117/229/10534.

The legal texts refer to German law and the GDPR. The Terms were last modified on 17 September 2026. A named company with an address and a tax number is better than most.

A recent edit date also suggests somebody still reads the documents, which I can't say for every app I'd review.

The site links to X at x.com/xeve_ai, Instagram at instagram.com/xeve.ai and Trustpilot at trustpilot.com/review/xeve.ai. No Discord or Reddit links were found. So there is no community space to ask other users questions before paying.

For a product with strict refund rules, I'd have liked a place where buyers compare notes in public.

Renewal and refunds

Subscriptions renew automatically unless cancelled before the next billing cycle. There are no refunds or prorations for early cancellation or unused time. One-time payments are non-refundable, and credits are non-refundable as well.

An address to write to is better than nothing, but it is a request, not a right.

EU users have a 14-day withdrawal right through a "Withdraw contract" button. That is the one real exit, and it comes from European law rather than generosity. Anyone outside the EU does not get it.

Credits expire one year after purchase, so a big pack bought on impulse has a shelf life. Combined with the no-refund rule, that makes the smallest pack the sensible first buy.

My plain reading: once you pay, the money is gone unless you are in the EU and act within the window. Treat every purchase like a non-refundable train ticket, and buy the shortest trip first.

Who owns what

Generated videos remain the property of Jay Web Development & Services. Users grant a worldwide, sublicensable licence to show generated media in the public feed, under the heading "User generated media". Media for the public feed is reviewed individually, anonymised and can be removed on request.

That is how a stock footage agency treats its own shoots, not how it treats a client's. In practice, most users will never care who owns a short clip of a character. Anyone who plans to use generated videos anywhere else should read that clause first.

User uploads are not accepted. I read that as a sensible guardrail. No uploads means nobody can feed a real person's photo into the generator, which closes the ugliest door in this whole category.

Your chats and your data

Conversations are processed only to run the service; they are not sold, not used for advertising and not used to train AI models. The named LLM providers are Together AI and DeepInfra. OpenAI rates a small random sample of conversations for quality.

ElevenLabs provides the voice. The Privacy Policy also names AtlasCloud for image and video generation. Naming every provider is more candour than this category usually offers.

It means a reader can look each one up, and it makes the no-training promise easier to check.

Chat sessions are deleted automatically 12 months after the last activity. Account data is kept until the account is deleted. Financial records are kept for the German retention periods, up to ten years.

The automatic deletion is the clause I like most. A year of silence and the chats are gone, without anyone having to remember to ask.

Records of blocked attempts to create abuse material are kept even after account deletion. That is the one exception to deletion, and I don't mind it at all.

My overall read: the privacy side is better than average, and the money side is harder than average. The company is careful with your chats and firm with your cash, and I'd plan around both.

The late discount

The first-payment discount appears only after some minutes. So someone who pays in the first rush pays full price, and someone who waits gets the cheaper first month. I'd call that a reward for hesitation, and my advice follows from it: wait.

The loose free limit

The site claims 5 free messages per character, but the limit was not enforced. That sounds like good news, and today it is. The problem is that a rule nobody enforces can snap shut without notice.

The busy character

While the video character is busy with an action, a notice reads "Voice input pauses while she is doing something". It is a small thing. In a product sold as a live call, a pause button on your own voice breaks the illusion more than a slow reply would.

The account nobody asked for

An affiliate account is created just by opening the affiliate page. Curiosity costs nothing here, but it does leave a trail. I think a click to look should not count as a click to join.

The content farm

The site has many SEO and comparison pages. Some of them rank the site against rivals, and the comparisons naturally favour the house. I'd read those the way I read a venue's own list of best venues.

It is not for users who want a large character library, or community-made bots. About twenty house faces are the whole cast. Once you have met them, there is nobody new.

It is not for users who want a companion without the adult side. There is no gentle mode to fall back on. And it is not for users who want refunds, because outside the EU withdrawal window there effectively are none.

My short version: curious adults with a month to spare, yes. Anyone looking for a long-term companion, a writing partner or a safe first purchase, no. If you are unsure which camp you are in, the free account will tell you within an evening.

A retention offer may appear while cancelling, with a "Cancel anyway" button. I'd read the offer, since a discount is a discount, and then press the button if the answer was already no. Retention offers are the "are you sure?" of billing.

Access stays until the end of the paid period after cancelling. So there is no reason to wait until the last day. I'd cancel right after paying for a monthly plan, keep the month, and decide again from zero.

That habit matters more here than on most apps. Because there are no refunds and no prorations, a forgotten renewal is simply lost money. Cancelling early turns the subscription into a one-time purchase, which is how I'd treat it anyway.

The statement

The statement descriptor is "CURJay Web Developm. / CCBill.com xeve.ai". That is worth knowing before the charge shows up. A charge from a web development company looks like fraud to anyone who forgot what they signed up for.

It also cuts the other way. A discreet descriptor keeps the app's name off a shared statement, which some buyers will count as a feature.

I'd wait until the paid period ends before deleting, so the month you paid for is not thrown away.

My checklist before paying: screenshot the price, note the renewal date, and know where My Membership lives. Then set a calendar reminder a few days before renewal, because the no-refund rule means late is expensive.

Browse the whole cast before paying. It is small, and if none of the faces works for you, nothing else on the site will.

Wait for the first-payment discount. It shows up after some minutes on the site, and it takes a meaningful bite out of the first month. Ignore the countdown timer completely.

If you pay, take one month, not a year and not the lifetime plan. Refunds are close to impossible, so the monthly plan is the cheapest way to be wrong. Cancel right after paying and decide again at the end.

Switch the chat model to High quality. The text replies are quick enough either way, and depth is the weak spot.

Test memory over two days, not two messages. If the character forgetting you overnight bothers you on the free account, it will bother you more on a paid one.

Spend the monthly credits on videos or images before they lapse. And untick the marketing checkbox at sign-up, because nobody needs more mail from a companion app.

The free tier is mixed. It shows the cast, a voice reply and a single image, but the video actions stay locked, and those are the reason to come.

Value for money is mixed. The flat price with unlimited video is simple and fair, and the no-refund rules and shifting prices take points off.

Images and voice are good, and this is the one area where the app earns its keep. The faces are consistent, the voice replies help, and the lip-sync is close but not clean.

Privacy and terms are mixed: careful with chats and named providers, firm with money and ownership. Cancelling and billing are mixed too, with a clear path but an unfamiliar statement name. Content settings are clear but offer no gentle mode.

It also loses on the library. A curated cast of about twenty faces cannot compete with apps where users build and share characters by the thousand. What it gains in consistency, it gives up in variety.

On price, it sits in the middle. The flat membership with unlimited video is easier to predict than token-heavy apps, and the refund rules are stricter than most. Predictable and unforgiving is an odd mix, but it is better than the reverse.

My rule of thumb: pick Xeve.ai for a video novelty, and a text-first app for a companion. They are different tools, like a gimbal and a tripod, and I'd rather own the right one than argue about which is better.

What it cannot do yet is keep the call interesting. The face is good, the voice is pleasant, and the conversation behind them never grows past a first date. The best thing it gave me was a story about a dog and his imaginary long fur, and I think that says as much about me as about the app.

My lasting thought is an editor's one. The better an app gets at faces, the more it matters what sits behind them. Here, the face is finished and the rest is still a rough cut.