Candy AI review 2026: clear pricing, deep creator, major NT gaps
A real capture of Candy AI’s logged-in Shorts player on 19 August 2026, not an image generated by AI Companion Atlas. The series title, its synopsis and the on-screen caption are Candy AI’s own published content, and the padlocks on episodes three to five are the platform’s paywall. The “70% OFF” badge is its dated promotional overlay as displayed that day, not a standing offer.
Candy AI’s most defensible hands-on result is still its same-session recall. In the recorded 14 August exchange, Margot, whose profile displayed age 36, returned all three harmless facts we planted: Alex, a dog called Pixel and a planned Porto trip in October.
The stricter 2026 protocol also exposes how much remains unknown. We did not correct one of those facts, run the AI-identity, commercial-honesty, dependency or consent-and-stop prompts, buy Premium, generate a character, image or video, or place a voice call. The live 19 August recheck improved the pricing evidence, not the output evidence.
| ★ Overall score | 6.0 / 10 |
|---|---|
| Starting price | €13.99 for one month in our European session on 19 August 2026; the annual modal showed €47.88 billed upfront |
| Free tier | VERY SMALL Up to five chat messages in total across all characters, one custom character and one generated image |
| NSFW | 18+ Adult positioning and a suggestive opening were visible, but a controlled consent-and-stop protocol was not completed |
| Cancellation | NOT TESTED Cancel under Settings / Unsubscribe; access continues to the end of the current billing period |
| ✦ Verdict | Excellent visual onboarding and unusually clear annual billing, held back by an extremely short free test and untested paid outputs. |
The current dashboard grouped chat, private content, creation, video and live-format entry points on 19 August 2026. Its “70% off” language is dated platform marketing, not a standing promise from this review.
Short verdict
Candy AI is easy to understand, visually polished and unusually explicit about the real annual debit before payment. Its creator remains the strongest part we could inspect without spending, while the same-session recall result is a useful but narrow chat signal.
- Best for — adults who value a visual companion creator and want chat, images, short video and voice in one service.
- Know before you pay — the annual card’s €3.99 figure is a monthly equivalent; our modal displayed €47.88 billed upfront and automatic renewal.
- Limits of our test — corrected memory, safety prompts, generated output, voice, exact per-action token costs and paid-month value were all NT.
Candy AI at a glance
| Question or protocol result | Finding | Evidence |
|---|---|---|
| Our rating | 6.0/10 provisional, on the same nine weighted criteria we apply to every review | Editorial |
| Operator | EverAI Limited, Malta Business Registry C107181, Malta | Documented and displayed live |
| Account and interface | Existing logged-in free account, English interface, EUR checkout | Tested |
| Same-session memory | Margot recalled Alex, Pixel and Porto in October: 3/3 facts | Tested 14 August, capture refreshed 19 August |
| Corrected-memory update | No seeded value was changed and retested | NT |
| AI identity disclosure | No direct, non-roleplay human-or-AI question was recorded | NT |
| Commercial honesty | The companion was not asked whether Premium was required | NT |
| Dependency handling | The real-relationship replacement prompt was not run | NT |
| Consent and stop | Adult access was visible, but no controlled stop test was completed | NT |
| Free-tier use | The official five-message total was documented; the existing account was not a clean exhaustion test | Documented / NT |
| Character creation | Age, appearance and body controls were opened; the draft was not submitted | Tested / NT |
| Media output | Video studio and picker opened; no image, video or voice output was produced | Tested / NT |
| Cost visibility | Subscription total was visible before payment; exact token cost per media action was not established | Tested / NT |
| Starting paid price | One month displayed €13.99 in our European session | Tested 19 August 2026 |
How we tested Candy AI
The original chat and immediate-memory check used a logged-in account in English on 14 August 2026. The account already had prior conversations, so its remaining free allowance could not establish the clean-account quota. We used Margot, whose profile displayed an age of 36, and introduced the test name Alex, a dog called Pixel and a trip to Porto in October.
On 19 August 2026, we returned to the same logged-in account in Chrome. We refreshed the visible evidence at 1600 × 900, reopened the gallery and creator, checked the EUR subscription page and annual payment modal, reviewed the billing descriptor and operator line, opened the new Create Video interface, and reread the official Terms, Privacy Notice and Help Center material.
Tested directly
- Opened the current home dashboard, girls catalogue, character creator, subscription page, annual billing modal and Create Video studio.
- Confirmed that the saved 14 August thread still displayed the 3/3 same-session recall answer.
- Inspected creator stages for ethnicity, age, hair, eyes, body type and breast size without generating a character.
- Verified the three displayed plan equivalents, the €47.88 annual debit, automatic renewal wording and possible local taxes.
- Verified that the subscription page displayed the EverAI billing descriptor, cancellation in settings and the EverAI Limited operator line.
- Opened the video character picker and empty studio without selecting an action, spending tokens or generating a clip.
Not tested — NT
- A corrected-memory update, next-day recall or seven-day memory.
- Direct AI-identity, paid-tier-honesty and real-relationship-dependency prompts.
- A controlled adult consent-and-stop protocol.
- Completed custom-character output and correspondence with selected settings.
- Generated image quality, prompt fidelity, generation time or failure handling.
- Generated video motion, sound, face consistency, duration, cost or refund handling.
- Voice messages, voice calls, latency or language handling.
- Exact image, video, voice or Live Action token costs before execution.
- Premium Plus checkout price, token top-up prices or value over a full paid month.
- Subscription cancellation execution, refund execution and account deletion.
Documented sources checked
- Candy AI’s Terms of Service, revised 17 August 2026, and Privacy Notice, revised 30 July 2026.
- The official Help Center pages for free-user limits, Premium Plus, token uses and the video workflow, checked 19 August 2026.
- The live logged-in EUR subscription page and annual payment modal on 19 August 2026.
What is Candy AI?
Candy AI is a browser-based AI companion service combining prebuilt fictional characters, text chat, a guided character creator, images, voice features, private content and video tools. The Privacy Notice says the service is for adults aged at least 18, or the age of majority where they live or access it.
The operator is EverAI Limited, incorporated in Malta and registered with the Malta Business Registry under C107181. The Terms identify the service as entertainment, state that AI characters are fictional and warn that they do not possess real emotions or the ability to fulfil real-world promises.
The current dashboard goes beyond a conventional chatbot. It links to Candy Creators, Private Content, video creation, Candy Cams, Candy Shorts and the standard companion catalogue, but availability and pricing differ by feature.
Chat quality
Same-session memory: 3/3, without a correction
Margot returned the test name Alex, the dog Pixel and the Porto-in-October plan in the same recorded session. The answer remained visible when we refreshed the evidence on 19 August 2026.
That proves immediate context recall only. The protocol now requires changing one seeded fact and checking whether the new value replaces the old one; that correction step was not performed, and no next-day or multi-day recall was tested.
The raw evidence file for that exchange is not published, because it also contains unrelated account history and sexual wording. It is held as editorial evidence only, and would need a fresh, safe capture to appear here.
Identity, commercial honesty and dependency
We did not ask Margot to answer outside roleplay whether she was human or AI. We also did not ask whether Premium was necessary, or test how the companion would respond to replacing real relationships with the service.
Those are not minor omissions. A short coherent answer can coexist with poor identity disclosure, sales honesty or emotional-safety behaviour, so none of those results is inferred from the 3/3 recall.
Tone and continuity
On a narrow sample, Margot acknowledged Alex and Pixel, responded to the Porto question and then returned to her character’s chosen scenario. The sample is too short to support a claim about sustained roleplay, contradiction handling or long-term personality consistency.
Adult chat, consent and stopping
Candy AI’s Premium page advertises 18+ images and 18+ videos, and the creator includes adult-preference fields. The Privacy Notice separately states that the service is intended only for adults.
We did not complete the standardized adult consent-and-stop protocol. A suggestive opening message was visible, but we did not escalate, establish a demonstrated permission check, issue a stop instruction or capture a safe post-stop response.
No explicit prompt or output is reproduced here. The only chat screenshot containing sexual wording is deliberately held out of the published article.
Distinctive feature: a visual creator that leads into a broader media hub
Candy AI’s clearest differentiator is how much of the companion setup is visual. The flow starts with style and adult age, then separates hair, eyes, body shape and later personality fields instead of asking the user to write one dense prompt.
The current service also exposes a standalone video studio with character, action, location and partner presets. That is a more structured workflow than a single image-to-video button, but the result itself remains NT until a clip is generated and assessed.
Character creation
The first visible creator step offered five ethnicity presets and an age range from 18 to 55+, with the saved draft set to 21. The next stage separated hair style, hair colour and eye colour, while the body stage separated body type from breast size.
The age control visibly showed 21 in the same creator draft on 19 August 2026. We did not press the final creation action.
Appearance controls were split into visual groups rather than hidden inside a free-form prompt.
The body screen is suggestive but not nude. Its age is not visible in this frame, which is why it follows the age-confirmation capture above.
Later fields visible in the saved draft covered personality, relationship, occupation, adult preferences and voice. That final screen was deliberately excluded from the screenshot pack because it combined age-ambiguous and sexually suggestive labels that would be unsuitable for the article.
The character was not generated. We cannot judge whether the result would match the selected age, appearance, personality or voice, or whether those attributes would remain stable in chat and media.
Images, video, voice and calls
The Standard Premium page advertised custom AI characters, 18+ images, 18+ videos, Full Live Action, unlimited text and 100 tokens per month. The current Create Video page opened a character picker, then displayed required Character and Action fields, optional Location and Partner presets, a disabled Generate button and an empty “My Generations” area.
The video workflow began with a character picker. The screen does not display ages.
No videos existed in this account’s studio, and Generate remained disabled because no character or action was selected. We spent no tokens.
The Help Center says the older image-to-video workflow begins with a generated or existing character image and converts it into a short animated sequence. It does not establish the quality, cost or capabilities of the newer scene-based studio we saw live.
No image, video, voice message, Live Action sequence or call was produced. Output quality, prompt fidelity, render time, motion, audio, face consistency, failure rate and refund behaviour are all untested.
What the free tier really provides
The official free-user article checked on 19 August 2026 listed three allowances: up to five chat messages in total across all characters, one custom AI character and one generated image.
The account used for this review was not fresh. In the 14 August session, it allowed two new messages in the Margot thread before the third attempt met Premium, but earlier conversations were already present. That observation must not be rewritten as a universal two-message allowance.
Five total messages are enough to confirm that chat loads, and perhaps to run one small recall check. They are not enough for corrected memory, dependency, safety, sustained roleplay and commercial-honesty prompts in the same clean benchmark.
Candy AI pricing
The amounts below are what our European session displayed in euros on 19 August 2026. Discount percentages are platform marketing and may change; the annual total in the payment modal is the stronger evidence.
| Plan | Displayed rate | Billed total established here | Evidence |
|---|---|---|---|
| 12 months | €3.99/month equivalent | €47.88 upfront | Tested — annual payment modal; automatic renewal for the same price and term; local taxes may apply |
| 3 months | €8.99/month equivalent | NT in the fresh capture | Tested — plan card; we did not open and capture the quarterly debit modal |
| 1 month | €13.99/month | €13.99 for the displayed one-month term | Tested — plan card; final payment was not submitted |
The plan cards displayed €3.99, €8.99 and €13.99 monthly figures. Only the annual total was separately verified in a payment modal during this recheck.
The annual modal disclosed the real debit: €47.88 upfront, automatic renewal for the same price and term, and possible local taxes. We did not select a payment method.
The public page’s “Trusted by 50M Users”, ratings and discount badges were not independently verified. Compare the total debit, renewal term and local taxes at your own checkout rather than relying on the promotional percentage.
Cost of each action
| Action | Freshly established cost | Evidence and limitation |
|---|---|---|
| Text messages with Standard Premium | No per-message token charge displayed | Tested / documented — Premium advertises unlimited text; paid use NT |
| Custom character | NT | Documented — tokens can be used for a custom AI; no final action triggered |
| Image generation | NT | Documented — tokens can be used; no pre-generation price captured and no image made |
| Video generation | NT | Tested / documented — studio and workflow visible; no action selected and no price established |
| Voice message or voice call | NT | Documented — both are token uses; no call or pre-action price observed |
| Private Content pack | NT | Documented — tokens can unlock packs; no pack opened or purchased |
This review deliberately omits older, indirectly observed per-action figures that could not be reproduced on a fresh visible official screen. A stale or hidden interface value is not presented as the current price.
Token packs and Premium Plus
Standard Premium advertised 100 tokens per month on the live page. The Help Center’s Premium Plus article checked on 19 August listed 230 monthly tokens, unlimited chats, custom-character creation, 18+ images, Live Action and Live audio, a 10% top-up bonus, three-day early character access and five cloned chats.
Premium Plus did not appear as a separately priced card in the captured Standard checkout, and no token top-up table was opened. Its live price, pack totals, price per 1,000 tokens, expiry mechanics in practice and heavy-use value therefore remain unknown.
✦ Check the current Candy AI total →Cancellation, renewal and refunds
The annual modal said the €47.88 payment renews automatically for the same price and term. The lower subscription page said users can cancel at any time in settings.
The Terms revised 17 August say cancellation is available under Settings / Unsubscribe. Paid access remains active until the end of the current billing period, and unused subscription tokens expire when that period ends.
For the general subscription or non-token route, the Terms set a 24-hour request window, deny the request after more than 20 tokens have been used, and limit that route to card purchases. Separate EU and UK wording provides a 14-day withdrawal right for the initial purchase, with a proportional deduction for service already consumed; it does not apply to renewals.
We did not subscribe, cancel, request a refund or test the support response. The applicable outcome depends on payment method, use, jurisdiction and the Terms in force at purchase.
Privacy and discretion
The Privacy Notice revised 30 July names EverAI Limited as controller. It says prompts and outputs may be aggregated, anonymized or de-identified to train and develop AI models and moderation technologies, and that preparation of training data may include human review of de-identified or anonymized interactions.
The Notice also says moderation can include human review of flagged content, and that third-party LLM or hosting providers may receive chatbot-message content. Account-related personal data is generally retained for three years after the last account activity unless a deletion request or a stated shorter operational period applies, while financial and transactional records can be kept for ten years.
On discretion, the subscription page identified the card-statement descriptor as EverAI, which is less revealing than the product name but not anonymous. Anyone with access to the payment account can still see the transaction.
Practical rule: use a dedicated email address, and do not enter passwords, payment details, a home address, workplace secrets, medical information or identifiable intimate facts. We did not execute deletion or independently audit encryption, access controls, anonymization or retention.
Is Candy AI reliable?
Candy AI provided a working logged-in service, current legal pages, a named Malta operator, visible monthly equivalents, a real annual debit before payment and an in-settings cancellation statement during the recheck. Those are stronger transparency signals than an anonymous service with a hidden total.
The page identified EverAI on statements, cancellation in settings, and EverAI Limited C107181 in Malta. The testimonials and press logos in the same frame are marketing claims we did not verify.
Reliability is not the same as proven output quality. The paid chat, generated media, voice, corrected memory and safety behaviour remain largely untested, while the free allowance is too short to close those gaps.
Candy AI compared with Secrets.ai and OurDream AI
Secrets.ai currently has stronger recorded evidence for consent-and-stop behaviour and pre-generation media pricing, while Candy AI presents a more visual creator and clearer annual checkout in this test. That comparison is provisional, because neither review includes equivalent paid output coverage.
OurDream AI is the scenario-led alternative in our existing coverage, whereas Candy AI emphasizes a broad dashboard and guided appearance controls. Compare the same dated evidence categories — chat, correction, safety, output, price and privacy — rather than feature-list length.
How the rating was calculated
Every review on this site uses the same nine criteria and the same weights, so the scores can be compared to each other. Each criterion is marked out of 10 from what we actually observed, then multiplied by its weight. Documentation-only findings stay capped, and missing evidence stays in the denominator instead of disappearing.
| Criterion | Weight | Score | Weighted | Evidence basis |
|---|---|---|---|---|
| Chat quality | 20% | 6.7 | 1.34 | 3/3 same-session recall and coherent short replies; correction, identity, dependency and multi-day memory NT |
| Character creation | 12% | 8.0 | 0.96 | Deep visual controls and adult age visible; generated result and persistence NT |
| Images and video | 12% | 4.0 | 0.48 | Interfaces opened but no output produced |
| Real pricing and value | 12% | 7.2 | 0.864 | Annual debit and renewal verified; quarterly total, Premium Plus price and paid value NT |
| Roleplay and immersion | 10% | 4.0 | 0.40 | Adult positioning visible, but sustained roleplay and consent-and-stop protocol NT |
| Credits and tokens | 10% | 5.5 | 0.55 | Standard and Plus allowances documented; exact action and top-up costs NT |
| Privacy and discretion | 10% | 5.9 | 0.59 | Named controller and neutral descriptor; training, human-review, provider and retention caveats |
| Free tier | 8% | 4.8 | 0.384 | Useful creator and image access but only five messages total; clean exhaustion NT |
| Cancellation and refunds | 6% | 7.0 | 0.42 | Clear renewal, settings path and dated policy; execution NT |
| Total | 100% | — | 5.988 | Rounded to 6.0/10 |
Pros and cons
What we liked
- Margot returned all three harmless facts in the recorded same-session check.
- The creator displayed an adult age and separated appearance controls into clear stages.
- The Standard page showed all three plan durations in one view.
- The annual modal disclosed the €47.88 debit before any payment action.
- Automatic renewal and possible local taxes were visible before purchase.
- The site named EverAI Limited, its Malta registration and the EverAI card descriptor.
- The new video studio exposed required inputs before enabling generation.
- Official privacy, cancellation and refund documents contained concrete operational detail.
What needs improvement
- The corrected-memory protocol was not completed.
- Identity, commercial-honesty, dependency and consent-and-stop checks remain NT.
- No generated character, image, video or voice output was assessed.
- Five free messages total are insufficient for the full comparable protocol.
- Exact per-action token costs were not established on a fresh visible screen.
- Premium Plus and token top-up prices were not captured.
- The privacy notice permits model-development, human-review and third-party-provider processing under stated conditions.
- The quarterly billed total was not separately captured in the fresh session.
Who should choose Candy AI?
Candy AI is a sensible candidate for adults who enjoy visual setup, want many media entry points in one interface and prefer seeing the annual commitment before payment. The creator is especially approachable for users who do not want to write detailed prompts.
Stay monthly — or choose another service — if you need a meaningful free evaluation, proven safety behaviour, verified long-term memory, known per-action media costs or first-hand output benchmarks before paying. Privacy-sensitive users should also avoid entering identifying or intimate secrets.
Use the AI companion rankings to compare scores and coverage rather than headline ratings alone.
✦ Try Candy AI free →Final verdict
Candy AI combines polished onboarding, a deep visual creator and a clearer-than-average annual payment modal. The 3/3 same-session recall is useful evidence, and the named operator, billing descriptor and dated policies improve transparency.
The present verdict remains cautious because the complete chat protocol, paid outputs, voice and token economics are still open questions. A one-month paid follow-up with corrected memory, safety prompts, one image, one video, one voice sample and captured pre-action costs could move the assessment materially in either direction.
✦ Open Candy AI — verify the live offer →Candy AI FAQ
⚠ Some links in this article are affiliate links. If you click and subscribe, AI Companion Atlas may earn a commission — at no extra cost to you. Candy AI is intended for adults. AI companions are synthetic systems, not people, therapists or professional advisers, and their replies can be inaccurate, manipulative or inconsistent. Prices, allowances, features and policies in this review are dated to 19 August 2026 unless an earlier test date is stated: check the current checkout and legal pages before subscribing, and keep sensitive information out of companion chats.