SecretDesires AI Review 2026: Pros, Cons, and Who Is It For

I want to start with a sentence the company wrote about itself, because it reframes everything else. In its own User Guide, the answer to what Hearts cost is this: "Exact costs are intentionally not listed here because they are set dynamically." It goes on to tell you "For current prices, limits, engine availability, and Heart costs, always use the values shown inside the app", and then, more bluntly, "Do not rely on an old screenshot, email, or guide for a price or quota. The amount and terms shown at checkout are authoritative for that purchase." There's a companion line about features varying "by account, plan, device, and active experiments", and a note that the platform "may show different plan presentations, promotions, limits, or feature experiments".

Normally I'd log that as a boilerplate hedge and move on. Then the proof turned up on its own. The plan panel screenshotted from a logged-in account shows four monthly cards and no billing-cycle toggle anywhere. No monthly, quarterly or yearly switch. Meanwhile the most detailed cost breakdown I read documents three cycles, with quarterly at 10% off and yearly at 20% off, and I checked all six of those numbers by hand. Two accounts, two different shops, one website. That's the dynamic-presentation disclosure demonstrating itself in front of me.

So read the rest of this review as arithmetic on a snapshot. I did the maths, I read the Terms, the Privacy Policy, the User Guide, the Security page and the Transparency page, and I went hunting through the arbitration section for something it promises and never delivers. What I can't do is promise you a price. Neither can they, and they say so.

It's an AI companion site where you build a partner from a guided creation flow, then chat with it, generate images of it, animate those images into short clips, and (this is the interesting part) actually call it. You must be over 18 to use it. There's a free tier plus three paid ones: Pro at $7.99, Ultra at $13.99 and Max at $19.99 a month on the panel I saw, and the paid tiers unlock better chat engines, better image engines and a monthly Heart allowance.

Hearts are the metered currency sitting on top of your subscription. Plan-included chat engines don't charge per message. The cutting-edge chat engines, the premium image engine, video, voice notes and calls all can. So the subscription buys you access and the Hearts buy you the good stuff, which is the bit most people miss until their balance is gone.

The company itself is split by where you live. The Terms say the agreement is with Playhouse Media LLC (8 The Green STE A, Kent County, Dover, DE 19901, United States) "or, if you reside in the European Union, between you and Playhouse Media Trading Ltd." (Lek Business Centre Level 1D, Triq L-Esportaturi Zone 2, Central Business District, Birkirkara, CBD 2040, Malta). Both addresses sit in the footer of every page, which is more than most of this category manages.

Here's the panel exactly as it appeared on a logged-in free account, header reading "Choose your plan. Unlock the full experience. No commitment, cancel anytime." Four monthly cards. Free at $0. Pro at $7.99 with Unlimited Messaging, image and video generation, phone and voice features, Advanced and Pro chat engines, 100 monthly Hearts. Ultra at $13.99 adding voice cloning and the Ultra engines with 200 Hearts. Max at $19.99 with the Max and Experimental engines and 300 Hearts, carrying the "Most Popular" badge. No billing-cycle switch anywhere in that modal.

The most detailed cost breakdown I read documents cycles that panel didn't offer: Pro at $7.99 monthly, $7.19 quarterly, $6.39 yearly. Ultra at $13.99, $12.59, $11.19. Max at $19.99, $17.99, $15.99. It says quarterly saves 10% and yearly saves 20%, and I ran all six. Every single one lands on exactly 10.0% or 20.0% off, to the cent. That's a clean, honest structure and I'll credit it. I'll also flag that it isn't what this account was shown, so treat it as something that exists for some accounts and check your own checkout.

Per Heart on monthly: Pro is $0.0799, Ultra $0.06995, Max $0.06663. So Max is 1.20 times better value per Heart than Pro and it's the cheapest per Heart of the three. Which means the "Most Popular" badge is sitting on the priciest card and on the best value per Heart simultaneously. Both halves are true, and that's rarer than you'd think. On the yearly rates it goes further, Pro at $0.0639 a Heart and Max at $0.0533.

And now the sentence that decides this whole purchase. Video is 720p in 5 or 10 second clips at roughly 50 and 100 Hearts. The workflow the platform itself recommends is generate a still on the premium engine, then animate it: 4 Hearts plus 100 Hearts, so 104 Hearts. Pro gives you 100 Hearts a month. The entry plan cannot complete the platform's own recommended workflow even once. Max's 300 does it twice with 92 left over. If video is any part of why you're here, Pro is not the tier. Top-up Heart packs do exist for buying more, though I'm not going to quote you a price for them because I haven't seen one I can stand behind.

Refunds: the Terms name their own policy a no-refund policy, all sales final, nothing for dissatisfaction, misunderstanding or technical issues. The one exception is a 14-day withdrawal period for EU residents. If you're in Italy, Spain, Germany or anywhere else in the EU, that 14 days is your entire answer, and it exists because you're contracting with the Maltese entity rather than the Delaware one.

Let me split this into what's genuinely impressive and what reading the documents turns up, because it's the strangest mix in this catalogue.

Who you're contracting with. Playhouse Media LLC in Delaware, or Playhouse Media Trading Ltd. in Malta if you live in the EU. Both addresses are printed in the footer of every page. Terms last updated 26 March 2026, Security and Transparency pages both 9 March 2026, Privacy Policy 24 October 2024. Hold on to that last date.

Refunds. The section is headed "Subscription and Refund Policy" and opens "Before subscribing, please read our no-refund policy carefully." It closes by having you acknowledge you agree to "this no-refund policy". The text: "All sales are final and non-refundable except for EU residents who are granted a 14-day withdrawal period." And "No refunds will be issued for dissatisfaction, misunderstanding, or technical issues." Four other brands in this catalogue run a 24-hour, usage-voided, card-only template. One runs 30 days including renewals. This is a flat no with a statutory 14 days for EU residents, so it's the harshest refund position here unless you happen to live in the right place.

The arbitration opt-out that isn't there. Near the top of the Terms there's a notice in capitals: "THESE TERMS CONTAIN AN ARBITRATION CLAUSE AND CLASS ACTION WAIVER", explaining you waive a judge or jury and waive class and representative actions. Same notice, next line: "You have the right to opt-out of arbitration as explained below." It is not explained below. I read the whole arbitration section looking for it. Seven subsections (Agreement to Arbitrate, Prohibition of Class and Representative Actions, Pre-Arbitration Dispute Resolution, Arbitration Procedures, Costs of Arbitration, Confidentiality, Severability) and not one of them describes an opt-out, a window, or an address to send one to. The document advertises an exit in capital letters and never tells you where the door is. For the record, arbitration runs under JAMS Streamlined Rules with hearings in San Francisco, California law governs, anything outside arbitration goes to Santa Clara County, proceedings are confidential, and claims must be filed within one year. EU residents contract with the Maltese entity and the same Terms grant them the 14-day withdrawal right. I'm stating those facts and not offering a view on how any of it plays out.

The security disclosure, which is the best here by a distance. AES-256 at rest via LUKS full-volume encryption with a unique key per customer project, backups encrypted with RSA-4096, TLS in transit. Cloudflare at the network layer with always-on unmetered DDoS mitigation across 330-plus data centres, a web application firewall, bot management, rate limiting and origin IP masking so the servers are never publicly exposed. Hosting on OVH, whose data centres are certified to ISO 27001, 27017, 27018 and 27701, with 24/7 physical security, video surveillance, badge access and mantraps. Authentication through Firebase (Google's service) with sign-in via Discord, Google or X, so the platform states it never stores your password. A SOC 2 Type I and II audit with Delve, started February 2026 and described as in progress, covering security, availability and confidentiality. Independent penetration testing by two named firms, Casco Security and BSK Security. And the concrete one everybody in this category wants: charges show up on your statement as "SDAI", with nothing identifying the nature of the service. Nothing else in this catalogue publishes anything close to that list.

Then the contradiction. The Security page, under a heading reading "Your conversations are not training data", says messages are "processed and immediately discarded, never stored, never logged, never used for training", with routing rules excluding any provider that trains on user data. Dated 9 March 2026. The Privacy Policy, the binding document, dated 24 October 2024, lists among the purposes it uses your information for: "Analyze, maintain, improve, modify, customize, and measure the Services, including to train our artificial intelligence/machine learning models." Those two cannot both be true as written. And be precise about the "never stored" half: it plainly describes the upstream model providers' retention, because the platform itself obviously stores your chats. You can reopen a conversation. Memory Notes and Auto Memory are documented features. So the promise on the marketing-facing page is narrower than it reads, and the older binding document reserves a training right the newer page denies. Anyone deciding what to type in there deserves both sentences.

The borrowed privacy policy. I compared ten long passages of the privacy policy against the privacy policy of another platform in this catalogue, passage by passage. Seven came back word for word identical. The three that didn't match differ only where the company name is substituted in, so the same security sentence appears with a different company as its subject. The matching seven include the entire three-sentence Popular Characters paragraph, the data retention paragraph, the de-identified data clause, the Do Not Track sentence and the model training clause. What that means: not that the two companies are connected. One is a Delaware LLC with a Maltese arm, the other a different Delaware corporation with a Cypriot arm, unrelated on paper. It means the privacy policy you agree to is boilerplate rather than a document written for this product, which is worth knowing when the security page sitting right beside it makes bespoke promises about this specific system.

The inherited Popular Characters clause, on its own terms. If a Character you create and set to Public passes a popularity threshold, they reserve the right to preserve its characteristics and keep it active even after you delete your data and your account, on the basis that it's holding conversations with many thousands of users. Their own mitigation advice is to keep information about yourself out of the creation flow for a Public character. Practical takeaway: keep your characters private unless you're happy for them to outlive your account.

Ownership. Both halves. The Terms are unusually direct that when you create a Character you "own all rights to that Character", and you own the text, images, audio or video it generates. You also grant a licence to use the Character and Generations as needed to operate and promote the Services, and on Content you submit the licence is "nonexclusive, worldwide, royalty-free, fully paid-up, transferable, sublicensable, perpetual, irrevocable" for any purpose related to the Services. Perpetual and irrevocable are the words to carry out of that.

Moderation and transparency. There's a dedicated Transparency page (9 March 2026) describing layered moderation: automated keyword filtering, AI classification scoring flagged items across violation categories on confidence thresholds rather than keyword matches, human reviewers for borderline cases, and third-party compliance verification. The prohibited list names child abuse material, unauthorised use of real people's likenesses, deepfakes of real people and non-consensual scenarios, among others. Reports go to a published address, and per the Terms, complaints about illegal content are committed to resolution within five business days with illegal content removed immediately on identification rather than at the end of an investigation. On legal requests, each government or law enforcement request is evaluated for legal validity, they say they push back on requests that are overly broad or legally insufficient, and they notify affected users when legally permitted. That last commitment is rare and I'll credit it. The page also says "Our first quantitative transparency report, covering moderation actions, legal requests, and safety metrics, publishes Q2 2026", with quarterly reports after. Q2 2026 ended in June. It's September, the sentence is still in future tense, the page hasn't been touched since March, the public compliance profile is still "coming soon" and the SOC 2 has been "in progress" since February. No outrage from me, just the date: they've committed further than anyone else here, naming auditors, penetration testers and a publishing schedule, and the first report hasn't landed.

Two more things that cost real money. Straight from the Guide: "Do not delete the account as a substitute for cancelling a subscription." Deleting your account does not stop the billing. Cancel first (Profile, Settings, Billing, Cancel Subscription), complete every confirmation step, keep the final confirmation. Deleting a Partner can also remove access to that conversation and its data, and the Guide says to treat it as permanent, while the Terms add that the company "is not responsible for the deletion or failure to store any data or content" and that long-inactive accounts may be terminated. Submitting a Partner to the public library is treated as irreversible too, later private edits don't update the public snapshot, and partners built on a face-swap reference or a cloned voice can't be shared at all. Finally, creating an account is consent to marketing email as well as transactional, and the Terms state "This consent is a material condition of your use of the Services", reserving the right to use your address for communications from "our affiliates and partners". You can unsubscribe from marketing through the link or your settings at any time, but not from transactional mail. Marketing consent framed as a material condition of using a service is unusual wording, and EU users contract with the Maltese entity. I'll leave the legal opinion to lawyers.

This is a product that publishes more about its security than anything else in this catalogue and less about its prices than anything else in this catalogue, and both of those are deliberate. The security page names its auditors, its penetration testers, its encryption, its ISO-certified hosting and the four letters that show up on your bank statement. The User Guide says the Heart costs are "intentionally not listed here because they are set dynamically" and then, sure enough, the panel from one logged-in account shows four monthly cards while the best cost breakdown I read documents three billing cycles. Two accounts, two different shops.

What I'd actually pay for here is the real-time calling and the face consistency. Those are real, they're hard, and hardly anyone else in this space does either well. What I'd go in braced for is the memory thinning out over long stretches, a Heart balance that a companion can draw down on its own initiative, and the fact that the platform's own recommended video workflow costs 104 Hearts while the entry tier hands you 100. And the paperwork: a Terms document that promises an arbitration opt-out in capital letters and never says where it is, and a privacy policy seven-tenths word for word identical to another platform's in this catalogue while the security page beside it makes bespoke promises. If you're in the EU you have a 14-day withdrawal window to change your mind. If you're not, the Terms call their own position a no-refund policy, so decide before you press the button.