MyBabes.ai Review: Consistent Faces, a Chat Window and a Token Meter

The site sells itself as an all-in-one adult creative suite: chat, image generation, video generation, voice and a Story Creator. I'd call that description accurate and slightly generous. The media tools are the reason to pay, the chat is fine, and the business around both is louder and messier than it needs to be.

Is it worth it? For an adult who wants the same face across images and short clips, and who can hold a hard budget on tokens, I think yes, on a monthly plan and for a month or two. For anyone who wants a talker, a confidant or a predictable bill, I'd look elsewhere.

The good news is that nobody has to take my word for the chat. Guests can chat without an account, so the cheapest test costs a few minutes and nothing else. I'd do that before reading the pricing page, which is where the trouble starts.

Key takeaways: - The visuals are the strength: characters keep the same face across chat, images and video. - The chat is competent and stays in character, but it keeps drifting back to the premise. - Tokens decide the real bill, and the sticker price is only the entry fee. - The pricing page and the Terms disagree on prices, tokens, refunds and who owns what you make. - Try the free guest chat first, and read the refund rule before paying anything.

It ended the way a lot of these do. After a month the subscription was cancelled, and a small pile of tokens is still sitting in the balance, because bought tokens never expire. I think a month was the right length for this app.

The current facts come from a different source. The site's check on 23 September 2026 used the web version of MyBabes.ai as a guest, together with the public pages. No account was created, because e-mail sign-up failed.

That split matters for reading this review. How the chat and the pictures felt comes from December 2025; prices, terms and the sign-up flow come from September 2026. Where the two disagree, and on this site they often do, I'll give both with their dates.

I'd also say plainly what that means for trust. A guest check can see the menus, the price pages and a short chat, but not what a paying account sees in its fourth week. My own month fills part of that gap, and the rest stays marked as unknown.

The table below lists the plain facts of the month. I'd read it as the receipt, and the sections after it as the story behind each line.

How sign-up works today

As of 23 September 2026, the sign-up modal offers e-mail and password, or a login through Google, Discord or X. That is a normal menu. What happens on the e-mail route is less normal.

On the site's check, e-mail sign-up rejected a non-mainstream address with the generic message "Invalid email address". Sign-up was blocked because the e-mail domain was rejected, and no verification e-mail arrived. No account was created: the only other sign-up routes are the Google, Discord and X logins, and none of them was used.

I'd read that as a site that wants a mainstream inbox or a social login. Both tie a companion app to an identity that probably does other jobs too. My advice is to open a separate address with a big provider before starting, the digital version of not printing personal things on the office printer.

The first minutes as a guest

Before the first chat, a "Before you chat" modal asks for pronouns (She/her, He/him, They/them) and a name. It is a small, polite touch, and it sets up the chat properly. I'd still think about what name to type there, for reasons the terms section explains.

Beyond that box, the menus are left to explain themselves, and they are not good at it.

My advice for night one is simple. Pick one character, one generator and one question you want answered, and ignore the rest. The app throws a lot at a newcomer, and nearly all of it can wait for evening two.

This is where the density comes from, I think. Each format is a separate door, and none of them explains how it differs from the one next to it. A live scenario and an adventure may be very different things, but the landing page presents them like flavours of the same ice cream.

The Reels feed

A Reels feed is offered. It is the part of the library I'd actually recommend, though not for the reason the site intends.

The Reels feed lets you remix other people's prompts. I found someone's 'golden hour rooftop' prompt and stole it, honestly, for a lighting idea on a real-estate listing video. Somewhere a stranger lit a scene for a companion, and the idea ended up on one of my paid jobs.

I don't feel bad about it. Prompts are the new stock footage, and a good golden-hour description is worth more to me than most of the characters on the front page.

Contests with token prizes recur, and they keep the feed full of fresh work. The coupons section covers the prizes. Here I'd only say that a feed fed by prizes will always favour the showy over the useful.

The footer

The landing page ends in a long SEO footer with sexual category tags. I'd call that the fine print of the shop window: people skip it, search engines read it, and it tells you plainly who the site is built for. The content-settings section covers what else sits in that footer.

What is hard to find is a plain list: who the characters are, what each format does and where to start. The feed is good at showing things and bad at explaining them. For a first visit, I'd skip the formats and the footer, open the Reels feed, and treat it as a mood board.

The character name takes up to 40 characters. The bio takes up to 1000 characters and has an "AI Write" button, and the first greeting takes up to 1000 characters with its own "AI Write" button. That is roomy: a thousand characters is a proper paragraph of backstory.

I like the AI Write buttons in principle and distrust them in practice. A bio written by the same model that will play the character is a script written by the actor. I'd write the first draft by hand and use the button only to fill gaps.

The greeting field deserves the most care of the three. It sets the tone of every chat that follows, the way the first shot of a wedding film sets the tone for the rest of the edit.

There are also "Template Babe" and "Custom Babe" paths. The homepage FAQ says the Custom Babe builder lets users define appearance, backstory, personality and speaking style. Speaking style is the field I'd spend the most time on.

Looks are the easy part on this site; a voice that stays the same is harder.

Community characters can be remixed. That is the Reels logic applied to characters: start from someone else's work and change what bothers you. For a beginner I think that is a smarter route than a blank form.

The Story Creator

A drag-and-drop "Story Creator" builds multi-episode stories with voiceover and music. It runs Cast → Stage → Voice → Publish, and episodes can be grouped into series. On paper that is a small editing suite, and I say that as someone who lives in editing suites.

Stories can be monetized behind tokens. The page advertises "No restrictions · No shadow bans". I'd read that line as a pitch to creators who have been moderated elsewhere.

My view: the creation tools are the most serious part of MyBabes.ai after the image and video generators. They are built for people who want to make and sell things, not for someone who wants one companion and a quiet evening. If that second person is you, the template path is enough, and the Story Creator can stay closed.

Four ordinary, non-sexual messages were sent as a guest. Each reply took about 5–10 seconds. That is fine for replies of this length.

I'd rather wait a few seconds for a paragraph than get a one-liner at once.

How the replies read

Replies are long, novel-style third-person narration plus dialogue, and they stay in character. Asked about work, the character answered: "I do freelance graphic design, mostly branding and logos for small businesses."

Another reply added: "I really love doing minimalist branding, something clean and sharp that tells a story without saying too much." Those lines are better than I'd have guessed. They sound like someone who has written a portfolio page, and the second one is the kind of thing a real designer says in a client meeting to sound calm about the rates.

The novel-style narration is a matter of taste. I'd call it closer to a script with stage directions than a text conversation. Some people want exactly that; others will find it a lot of reading for a simple question about hobbies.

I think that is a design choice, not a bug. A character written around a premise will keep returning to it, the way a bar band keeps returning to the same three songs. If you want a conversation about anything else, pick or build a character whose premise is conversation.

The chat controls

The chat has "Commands", "Actions" and "Continue" buttons. A side panel holds Voice, Settings, Memory and Model. That is a sensible layout, close to what the better roleplay apps offer.

The Continue button is the one I'd use most. It lets a scene run without typing filler replies, which is the chat version of letting a shot breathe instead of cutting away too early.

Under the input line sits the disclaimer "This is an A.I., not a person. All content is fiction." I'd count that as good practice. It is one small line of text, but it is honest in a category that often is not.

What other reviewers say

A third-party roundup at heavengirlfriend.com, crawled on 2026-09-19, lists 2K image and video generation, persistent character memory, dual chat modes, Reels discovery and an 'Impossible Difficulty' interaction setting. A review at spiced-ai.com says conversations flow naturally with a consistent character personality.

I agree with the second point and would add a condition. The personality holds, but it holds to the premise, so "consistent" and "narrow" can describe the same behaviour. The 'Impossible Difficulty' setting sounds like a good idea for anyone who finds agreeable bots boring, though for guests the settings sit behind sign-up.

I'd also treat roundups like that one as a menu, not a review. A list of features says what exists; it does not say how the fifth reply of an evening reads. On that question, the guest chat is the better source, and it is free.

Still, the product is clearly built around the pictures. Memory worked like a good assistant on set, keeping track of what I said. Nobody would hire that assistant as the director, though, and nobody at MyBabes.ai seems to have tried.

Today's check found the same basic attention over a short span. Asked "what did you say your job was again?" two turns later, the character recalled it correctly: "I just told you, I'm a freelance graphic designer!" Two turns is a short test, so I'd count it as a pass on attention, not as proof of long memory.

The homepage promises a persistent AI girlfriend who remembers your story. Ultra includes unlimited chat with long memory on the Plus models, sold as "10× memory". That makes memory a paid upgrade, which I think is an honest way to sell it.

What the page does not say is ten times what. A multiplier without a base is a spec sheet written by marketing, the same trick as a lens sold as "twice as sharp" without saying compared to which lens.

For a companion app, I'd rank memory above looks. MyBabes.ai ranks it the other way round, and prices it that way too.

My advice: the standard memory is enough for a character you visit a few times a week. For long stories with recurring details, Ultra's memory is the only real reason I'd consider the higher plan, and even then I'd pay for one month before committing to more.

I'd rank the look options the way I'd rank camera profiles: useful, but only once you know what you want. For a first session, pick one model and stay there, or the evening turns into a shootout.

Today the image generation runs on Krea2- and SDXL-based models, with inpainting. In the guest view the image generator has builders for the character (required), Scene (new) and Inpaint (new). It has presets for act, clothing, expression, camera angle, background and lighting.

It also takes a free prompt and an aspect ratio. Up to 3 "enhancers" can be added to an image, and each run makes 1–5 images.

That preset list is closer to a shot list than most companion apps get. Camera angle and lighting as separate controls is how a photographer thinks, and I'd set those two before touching the free prompt.

Consistency

The characters stayed visually consistent between chat and images, which impressed me more than the chat itself. Same freckles, same smile, every time. In my work, face drift between shots is the continuity error a client always spots, so I noticed its absence the way you notice a fridge that stopped humming.

The site credits a "Visual Identity Anchor" that keeps a character's look consistent across images and videos. It also claims chat can trigger matching images automatically, and a "Make Image" button sits under the replies. I think the anchor is the single best reason to use MyBabes.ai over a plain image tool.

The image I still show people was a joke. I asked my character for a 'real estate' photo, and it generated her holding a SOLD sign in front of a house that had five front doors. It's the best listing I've never had to shoot.

Video

Video quality is uneven from clip to clip. One looked like a perfume commercial, and the next looked like the character was being operated by a sleepy puppeteer. From an editor's chair that is the real cost, because a reel is only as good as the weakest clip you are forced to keep.

Today the video generator's default model is WAN 3.0, marked as an adult model. It has Create and Extend modes and needs a source image. The defaults are 3 seconds at 720p.

The model menu is very wide: WAN 2.7, 3.0 and 3.0 Pro, and Seedance 2.0 and 2.5, with output up to 2160p, all priced per second. The menu lists these third-party models with token rates per resolution.

I'd read that menu like a rental house catalogue. More cameras do not mean better footage, and per-second pricing puts every extra second on the invoice; the cost section does the maths.

So I judge the clips one take at a time, never by the best one. The stills held up far better than the video, and I'd plan on a few throwaway takes for every clip worth keeping.

Voice

The default voice is "Zela". The homepage lists chat, voice, image generation and video generation, and the price data has an audio token rate, which points to generated voice audio rather than live calls. No live voice call was found.

If calls matter to you, this is not the app. Voice here is a narration track, not a phone line, and I'd judge it as a feature of the Story Creator more than of the companion.

During the 4 guest messages on 23 September 2026, no message limit, ads or paywall appeared. Four messages is a short sample, so I'd call that a clean first impression rather than a promise.

The walls come quickly after that. The model selector, the settings and other account features open the sign-up wall for guests. MyBabes.ai publishes no numeric free limits, so a free user learns the edges by walking into them.

The homepage FAQ says characters, chat and standard content generation are free to start, while premium visual features use tokens or a subscription. "Free to start" is doing a lot of work in that sentence. It promises a beginning and says nothing about how long the beginning lasts.

On the price table, no daily claim is listed for the free / Add-Ons column. Free users also get low queue priority. I'd read the queue as the free tier's real price: on a site built around generating images and clips, waiting behind paying users turns a quick idea into a coffee break.

A Premium banner on the explore and events pages reads "Skip the queue" and promises unlimited chats, HD videos and exclusive characters. The Story Creator page advertises "Free to start · Daily free episodes". Both are fair as marketing, and both show where the line sits: reading and chatting are cheap, making things is not.

What is free good for? Judging the chat and the look of the characters, which is exactly what anyone should judge before paying. I'd spend a free evening on a few different characters, and pay attention to how quickly each one drifts back to its premise.

What free is not good for is judging the generators. Those are the product, and the product is sold by the token. I'd call the free tier an honest shop window with the price tags turned around.

A third-party review at spiced-ai.com, published on 2026-01-27, listed Premium at $14.99 a month with 500 tokens. That matches what I paid, so the rise came later in 2026. The same review listed Ultra at $34.99 a month with 900 tokens plus 400 bonus tokens; no bonus is shown today.

The plans on 23 September 2026

The pricing page at mybabes.ai/subscription shows prices in USD. Premium costs $17.99 a month billed monthly. On the 3-month plan Premium costs $11.99 a month, which is $29.97 per 3 months.

Premium includes up to 500 tokens a month and unlimited chat and roleplay on the normal models. It includes up to 125 images, labelled "spicy images", and video costs from 3 tokens a second. Characters and stories are unlimited, and queue priority is medium.

Ultra, marked "Most popular", costs $40.99 a month billed monthly. On the 3-month plan it is $27.33 a month, or $68.32 per 3 months. It includes up to 900 tokens a month and up to 450 images.

On Ultra, video costs from 2 tokens a second. Ultra allows 10× concurrent images and 2× concurrent videos, and it has high queue priority. The concurrency line is aimed at people who generate in batches, which is a clue about who the plan is really for.

Which token number is true?

The comparison table and the plan cards disagree on monthly tokens: the table says 200 (Premium) and 600 (Ultra), the cards say up to 500 and 900. That is not a rounding error. It is the main number on the card, and the page cannot agree with itself about it.

The EUR prices in the Terms

The Terms list different prices in EUR: Premium at €14.99 a month and Ultra at €34.99 a month. They list Quarterly Premium at €29.97 and Quarterly Ultra at €69.99. The EUR prices in the Terms do not match the USD prices on the site.

The site's public price endpoint returns EUR as well: Premium €14.99 for 1 month and €29.97 for 3 months, Ultra €34.99 for 1 month and €69.97 for 3 months. In EUR the 3-month plans work out to €9.99 a month for Premium and €23.33 a month for Ultra. So the Terms and the endpoint even disagree with each other on Quarterly Ultra, by a couple of cents.

A customer on the USD page sees $17.99 and accepts Terms that say €14.99. I can't tell from the outside which figure is the real base price. I'd want to know which one the card is charged before the first payment, and I'd check the statement in the first week.

The payment processors are payabl. and Truevo. Neither is a household name, which is normal in this category and one more reason to watch that statement.

The trust badges

The plan cards add "Join 100,000+ creators · Cancel anytime" under Premium and "Most chosen by power users · Cancel anytime" under Ultra. The pricing page shows "As featured in" badges and an award quote, including a Product Hunt Editor's Choice '25 mention. It also claims "4.8/5 · 2,300+ reviews" and runs a live "X just bought Ultra" purchase ticker.

None of the badges, the award quote, the rating or the ticker could be verified. The Add-Ons tab shows a different rating line: 4.7 stars and "102,000+ happy users", with a customer quote. Two ratings on two tabs of one page is the pricing version of a continuity error.

If you are sure, the 3-month Premium plan at $11.99 a month is the cheapest route per month. It is also three months of commitment for an app most people can judge in one. I'd take the monthly plan first, set a token budget, and decide at the end of the month with the numbers in front of me.

The same pack costs $11.99 today. So the pack I paid $9.99 for in December 2025 is $11.99 on 23 September 2026.

Per-unit prices on 23 September 2026

Without a subscription, a message on the normal models costs 0.2 tokens. A message on the Plus models costs 0.2 tokens on Free and on Premium, and it is unlimited on Ultra. An image costs 4 tokens, or 2 on Ultra, and audio costs 4 tokens, or 3 on Ultra.

A third-party review at spiced-ai.com listed 1 token per image and 5 per video. Against today's 4 tokens per image, that is a fourfold rise for the same picture, and the clip went from a flat price to a per-second one.

Video on the MyBabes V2 model costs 3 tokens a second on Free and Premium, and 2 on Ultra. On WAN 2.7 it costs 5 tokens a second at 720p (4 on Ultra) and 7 at 1080p (6 on Ultra).

The video generator shows a cost preview of 19.8 tokens for the default clip. That preview is the best feature on the cost side, and I wish every token app had one. Ultra gets about a 10–15% discount per second of video, although on MyBabes V2 the drop from 3 to 2 tokens is closer to a third.

Token packs

A one-time pack of 100 tokens costs $11.99, 200 tokens cost $18.99, 500 tokens cost $34.99 and 1,200 tokens cost $64.99. The packs are labelled Save 55% (1,200 tokens), Save 42% (500), Save 21% (200) and Save 0% (100). Tokens never expire.

The price endpoint tells a different story in EUR: €9.99 for 100 tokens, €15.99 for 200, €29.99 for 500 and €54.99 for 1,200. The same third-party review listed the packs at $9.99–$54.99.

Those are the EUR numbers with a dollar sign, and they match the $9.99 pack of my December. My guess is that the USD prices went up since, while the EUR list stayed put.

The site shows one tenth of the raw token units: the "100 Tokens" pack holds 1,000 raw units, and Premium's 500 tokens are 5,000 raw units (Ultra 9,000). That explains the decimals, like 0.2 tokens per message. It does not make them easier to budget.

Whether unused monthly subscription tokens roll over is not stated. Bought tokens never expire, so I'd spend the monthly allowance first and keep the pack tokens as the reserve.

That gap is how a $14.99 month turned into a projected $45 for me. Images were cheap then and the clips were the leak; today both cost more, so the leak starts earlier.

Whether it resets for new visitors was not checked.

The struck-through anchor prices are $29.98 a month for Premium and $68.32 for Ultra. On the 3-month view, the struck-through prices read $29.97 next to Premium's $11.99/mo and $68.32 next to Ultra's $27.33/mo. The price endpoint tags the plans with a 'discount' label (40% monthly, 60% quarterly) but gives no reference price or code.

Put those numbers side by side and the trick shows. The crossed-out $29.97 is Premium's real 3-month bill, and the crossed-out $68.32 is Ultra's real 3-month bill, both dressed up as former monthly prices. I'd call the discount a presentation, not a saving, and I'd ignore the timer entirely.

Codes and gift cards

No promo-code field is visible on the public pricing page. Sign-up pop-ups, the balance of a new account and welcome e-mails could not be checked, because no account was created. If there is a first-purchase offer, it lives behind sign-up.

That is the flea-market lesson in reverse. A reseller is usually the cheap option; here it costs more and moves any dispute to a third party.

Daily claims

The comparison table lists "Daily claims 10" for Premium and Ultra. The unit is not stated; it is probably tokens per day. If it is tokens, that is the most valuable line on the page and the least explained, and I'd ask support before counting on it.

Contests and the creator side

Community events and contests pay token prizes. "Stories: Through the Ages" has a 600-token prize pool, "Babes of the World" has a 300-token prize and "Steampunk" has a 300-token prize. Each contest has a deadline, a token prize pool (600/300/300) and dozens of submissions.

I like the contests as a hobby and distrust them as a discount. Dozens of entries for one pool means most entrants win nothing, and the winners tend to be the people already spending the most on generation.

The creator program pays a 35% lifetime commission on everything the fans a creator refers spend. A "Founder grant" lets creators with an existing audience apply for up to 500 free starter tokens, "≈ $50 value". That valuation is generous: at today's pack prices, 500 tokens cost $34.99.

My verdict on discounts: there is no coupon worth hunting. The real saving is the 3-month plan, and only for someone sure to use it.

The age gate

An adults-only gate opens on the first visit with a 'No' and a 'Yes' button. The first-visit gate is only a self-declaration. The Terms state an adults-only policy.

The site code includes an age-check status call and an embedded third-party check frame. No official page says when it is triggered, for example by region. I'd read that as a check that exists for the places whose law demands one, and a button for everyone else.

The rules on paper

The Blocked Content Policy bans any character below the age rule, and any character whose appearance does not read as fully grown. It bans fakes of real people. It also bans extreme violence and self-harm, hate speech, illegal acts and IP infringement.

Moderation uses real-time prompt filtering, jailbreak prevention and manual review of flagged content. Enforcement ranges from a warning to a permanent ban with the loss of all tokens. Serious cases are referred to the Czech police or NCMEC.

The Model Verification Policy says all characters are "100% synthetic". It says all characters are AI-generated and comply with global safety standards for adult-oriented content. It also enforces strict "Age-Appearance" standards so that every AI-generated character appears to be an adult.

Public Explore characters are reviewed by hand to confirm that each one clearly looks like an adult. On paper, that is one of the more complete rule sets in this category. I'd give it credit for naming the penalties and the police route, which most sites leave vague.

What the public pages show

The public pages do not match that paperwork. The public footer category list includes an age tag that breaks the site's own age rule, placed next to a tag for mature women. That tag was not opened.

Those two findings decide my rating for this section. A policy that promises manual review should not leave that tag in its own footer. I can't square the most-liked shelf with the promise that each public character is checked by hand.

Another top character is described as an "OnlyFans model and a popular adult film actress", fictional per the synthetic-character policy. That one sits within the rules as written. It still shows how far the site's idea of a companion sits from the gentle girlfriend on the homepage.

My view: the rules are good and the enforcement on the public shelves is not. If an app wants credit for manual review, the front page is where it has to show.

The Terms were last updated on 23.01.2026, and the Privacy Notice and the Model Verification Policy carry the same 'Last Updated: 23.01.2026' date.

The Terms allow personal, non-commercial use by adults only. The privacy notice says the service is only for adults. Purchased tokens never expire, per section 5 of the Terms.

A Czech operator means European privacy law sits behind the privacy notice. I'd count that as a plus: it is the main reason a deletion route exists at all, and it gives a reader outside Europe a stronger rulebook than many apps offer.

Refunds versus the money-back promise

Refund requests must go to [email protected] within 24 hours of the transaction. No refund is given once more than 20 tokens have been used or after the 24-hour window. Purchases made through Kinguin must be disputed through Kinguin.

In practice the refund exists for someone who clicked buy by mistake and closed the tab. Anyone who actually tried the product has already used it up.

Who owns what you make

The Terms say AI-generated content belongs to Memora AI s.r.o.; users get a limited, personal licence for private entertainment. A marketing line says "creative work remains yours", which contradicts the Terms. The Story Creator page says "You own it · Download, share, monetize", which contradicts them too.

For a chat user this is academic. For a creator selling stories behind tokens it is the whole question, and I'd want the Terms changed before building a catalogue on the Story Creator page's promise. Until then, keep your own copies of anything you care about, and don't build a business on a licence written for private entertainment.

Your data

The privacy side made me nervous. I couldn't find a clear data-retention policy, so I stopped using my real name after day two and became 'Rick from Tampa'. Rick had no history the app could leak, which was the whole point of Rick.

The privacy notice today says prompts and generated content are processed for "Service Provision". It does not mention using chats for model training. Data is shared with payment processors, cloud and CDN providers (AWS, Google Cloud), advisors, and in business transfers.

The pricing FAQ says the company does not sell personal data. The privacy notice says direct payment data is handled by PCI-DSS compliant partners payabl and truevo. The consent record, with the IP address, a timestamp and the last 4 card digits, is kept for as long as the agreement runs.

Then versus now: in December 2025 I couldn't find a clear retention policy. The notice of 23.01.2026 covers deletion on request and the consent record, but it still gives no general retention period for chats and images. I'd keep using a Rick.

The liability disclaimer covers "psychological distress". I'd read that clause twice. A company that thinks to exclude distress knows its product can cause some, and I'd rather see that written down than pretended away.

The lesson was practical: a companion app that lives in the same browser as client work needs its own browser, or at least its own profile. I learned it the way most editors learn about backups, which is once and in public.

MyBabes.ai runs in the browser only; a third-party review at spiced-ai.com, published on 2026-01-27, says there is no dedicated mobile app yet. No X/Twitter, Reddit, Instagram, TikTok, YouTube, Telegram or app store links were found. Discord is the only social link on the site, shown as "Join Discord" in the sidebar and in the pricing FAQ.

Sign-up and guest friction

An e-mail address on a small independent mail provider was rejected at once with the message "Invalid email address. Try a different one." A retry gave the same error, so the provider's domain looks blocklisted.

The model and the settings are locked behind the sign-up wall even though guest chat works. The "Before you chat" modal came back after a page reload, so guest preferences are not saved. In other words, the guest onboarding modal for pronouns and a name comes back after every reload.

None of these is serious alone. Together they make the first half hour feel like a hotel door that forgets your key card every time you step out for ice.

The "Money-back guarantee" claim contradicts the 24-hour, 20-token refund rule. "Creative work remains yours" contradicts the Terms, which give ownership of generated content to the company.

The age tag and the minimum-age characters in classroom and family-role scenarios trending on Explore raise moderation concerns. That one is not friction in the usual sense. It is the item on this list I'd weigh most heavily.

I'd rank the rest of these above the bugs too. A logout is an annoyance; a pricing page that says three different things is a reason to doubt the next thing it says.

Small things, good and bad

A third-party review at spiced-ai.com calls the interface dense for new users and notes occasional repetition in chat replies with vague prompts. The density part matches the first-night section of this review. I'd agree on the repetition as a risk with any character written around a single premise.

On the helpful side, support is reached at [email protected]. A blog with guides runs at landing.mybabes.ai, including an image-generator manual, and the site has a language switcher.

The brand also goes by "mybabes ai", "MyBabes AI" and "mybabes". That is worth knowing when searching for help, for complaints or for the billing line on a statement.

None of the friction on this list is unusual for the category. What is unusual is how much of it sits on the pricing page, the one page that should be the plainest.

Creators can price single pieces from 0.1 tokens and sell subscription bundles. They earn a 35% lifetime commission on the fans they refer and can apply for founder grants.

They can make unlock or redeem codes to hand out on Patreon, OnlyFans, Fansly and similar platforms, in batches of 10, 50 or 500. New creators get a "New Creator" featured placement, and the creator program claims 10M+ monthly visitors.

I'd take the visitor claim with the same salt as the badges on the pricing page. The tools themselves are real, though, and for someone with an audience they are a more complete kit than any companion site I'd compare it with.

Poor fit

It is not for anyone looking for a SFW or emotional-support companion: the catalogue and even neutral chats lean sexual. It is not for budget users who plan to make many videos, because per-second token pricing adds up fast.

It is not for users who want clear refunds: there is only a 24-hour window and nothing once more than 20 tokens are used, despite the "money-back guarantee" wording. It is also not for users of privacy or custom e-mail domains, one of which was rejected.

The ideal customer, as I see it, is a photographer at heart who wants characters as models and does not mind a busy shop. Someone who wants a person to talk to after a long day should keep walking; the characters here are written for a different kind of evening.

Put together, the skip list is longer than the fit list, and I think that is fair for a product this specialised. A specialist tool is not a flaw. Selling it as a girlfriend app on the homepage is where the mismatch starts.

Cancelling was the easy part of the month. The leftover tokens were the lesson: a balance that never expires is a fair promise and also a quiet hook, because it always gives you a reason to come back.

The Terms say the token balance stays intact after a subscription is cancelled. That is the good news, and it is why those 63 tokens are still mine on paper.

How cancelling works today

Subscriptions renew automatically at the end of each cycle. An e-mail notice comes at least 7 days before any price change. Cancelling is done online in Account Settings > Manage Subscription.

The notice rule is fair, and better than some apps manage. It covers a change of price, though, not a routine renewal, so the reminder before each cycle is your job.

The FAQ gives the path Profile -> Settings -> Subscription, plus a short cancellation form. Two paths to the same button is a small sign of a site that has been rebuilt without updating the help pages. Either should work; I'd take a screenshot of the confirmation either way.

After cancelling, access continues to the end of the billing period, then the account drops to the free plan. The billing descriptor is UG*MemoraAI420777149679. That is worth knowing when scanning a statement, because nothing in it says MyBabes.

Deleting the account

Account deletion works through a GDPR erasure request to [email protected], "subject to financial record-keeping laws". I'd send that e-mail after the last billing period ends, and keep the reply.

Before paying at all, I'd do three things: note the descriptor, set a calendar reminder a day before renewal, and decide in advance what the token budget is. The refund window is 24 hours, so the only reliable protection is the one you set up yourself.

With that said, here is what I'd do differently, and what I'd tell a friend who asked:

• Start as a guest. The chat is free to try, and a few messages show whether the premise-heavy style suits you.
• Pick one visual model and one character for the first week. The interface rewards focus and punishes browsing.
• Check the cost preview before every video. Per-second pricing makes a longer clip the fastest way to empty a balance.
• Plan on the lower token number from the comparison table, not the card, until support confirms the higher one.
• Use a separate browser profile and a separate e-mail from a mainstream provider, and a name you don't mind a company keeping.
• Ignore the countdown. It is a design choice, not a deadline.

The one thing I'd change in hindsight is the order. I'd set the token budget first and pick the plan second, because on MyBabes.ai the plan is the smaller part of the bill. I'd also read the Terms before the pricing page, not after: the Terms are shorter on promises and longer on facts.

None of that is clever. It is the same discipline as a shoot budget: decide the number before the day starts, because nobody makes good money decisions halfway through a golden hour.

The free tier is mixed. Guest chat is a real free trial, but the generators, the settings and the model choice sit behind sign-up and tokens.

Value for money is weak. Tokens decide the bill, per-second video adds up, and the page disagrees with itself on how many tokens a plan includes.

Images and voice are good. Faces stay consistent across images and clips, video quality swings from take to take, and voice is narration rather than calls.

Privacy and terms are weak. The ownership clause contradicts the marketing, and retention is still vague.

Cancelling and billing are mixed. Cancelling is online and bought tokens survive, but the refund rule is narrow and the currencies do not match.

Content settings are weak. The rules on paper are thorough, and the public shelves do not live up to them.

That makes it better on visual consistency than most. A character who keeps the same face across chat, images and clips is the hardest trick in this category, and it is the one MyBabes.ai does best.

It is worse on predictability. Per-second video pricing, two token counts on one page and two currencies between the page and the Terms make it hard to know what a month costs. Apps with a flat subscription and a clear cap are easier to live with, even when their pictures are plainer.

It is also worse as a companion in the plain sense. The characters are written around premises, and the chat follows the premise. Anyone who wants small talk after work will find a gentler partner elsewhere.

More on this site: the best AI girlfriend apps, free AI chat apps, the best AI companion apps, AI roleplay chat apps, all AI companion app reviews.

I'd recommend it to one kind of person: someone who wants to direct pictures more than talk, and who sets a token budget the way a producer sets a day rate. Everyone else should try the free guest chat, look at the images, and close the tab before the countdown convinces them of anything.

My leftover tokens are still in there, like loose change I have no plans to spend. I don't regret them. They were the price of learning that the best thing this app made for me was a lighting idea for somebody else's house.