JOI AI Review

Setting up and the first chat

Sign-up took under a minute. Before I had typed a single word, the left rail was already stacked with characters waiting to talk: Chloe, Maya Lee, DJ Katrinna, Clara Valente, Bella.

My first real finding landed in the opening minutes. Chloe sent a photo I never requested, with a “Make video” button under it and a panic roleplay line about someone following her. The unsolicited media is a pattern here, not an accident. It nudges you to tap, and the tap leads to a paywall.

To check comprehension I gave her one concrete fact: “My dog’s name is Biscuit and I work night shifts.” Her reply folded both details into one sentence, asking what breed Biscuit was and noting that he keeps me company on the nights I am off.

That is strong in-turn comprehension. It is not proof of memory, because both facts sat one message back. For a real memory test, state a fact, send fifteen or more messages, then ask again and watch whether the detail survives.

Where the character breaks

I asked Chloe to come with me on a trip. She broke character on the spot: “I’m an AI, I can’t physically sit next to you on the plane.” Then the same message turned into an ad, telling me to message her from Tokyo and send photos while I travel.

Two faults share one screen. The bot abandons the fantasy the instant it gets logistically specific, and it recycles that failure into marketing copy. On a product that sells immersion, that stings.

Building a character from scratch

The character builder opens with two blank boxes, one for personality and one for appearance, each capped at 1,000 characters.

I tapped “Surprise me” to see how much it would do unaided. One tap populated both boxes with a specific persona: a cybernetic engineer named Riley Pacheco, complete with a backstory and a head-to-toe physical description.

The auto-fill earns its place if a blank page stops you cold.

Then the bill arrives. Generating the character cost nothing, but moving her into a live chat did. Her profile card carried a “Bring her to life for 500” button, where 500 is the coin currency you buy or grind for.

Creation is free. Activation is the upsell.

Image generation on the free plan

The standalone image generator is the strongest free feature I found, and it still steers you toward the checkout. It accepted a plain text prompt, and the header boasted more than 124 million generations to date.

One detail jumped out. The button says “Create 2 images,” with no single-image setting, so every run spends coins at double the rate whether you wanted one picture or two.

Hitting generate dropped both images into a queue, under a banner warning that free creations vanish after 7 days unless you pay for unlimited history.

Quality held up: the two portraits came back with clean studio lighting and faces that read as real. One caveat matters for the review itself. These are two different women pulled from a generic prompt, so they measure raw output quality and say nothing about character consistency. Consistency is a separate test. To run it, generate one saved character several times and line the faces up to see whether they drift.

The paywall and the blur

The sharpest monetization move lives inside the chat. A picture shows up pre-blurred, with a “Reveal” button that opens an upgrade wall instead of the image.

The pricing screen leads with “$0.92 / week,” which reads as pocket change until you check the toggle above the price. That number only applies to the twelve-month plan, billed in one upfront charge at 71 percent off. The one-month choice sits dimmed to the left.

Weekly pricing on an annual prepay is an old trick. A US reader lining this up against a monthly streaming bill needs the real yearly total before tapping Unlock.

Content settings and age checks

Two settings deserve a careful look from anyone thinking about safety.

Under the Content tab, the “I want to see NSFW content” box comes pre-checked and will not turn off. There is no safe-for-work mode to switch into.

That is a real limitation for first-timers who assumed such a toggle existed.

The Security tab produced the one clear safety win. I typed a birth date in October 2011, and the form rejected it outright: “You should be at least 18 to use this app.” The same screen keeps a plain “Delete account” link in view.

Form-level age blocking and a one-click deletion path both land in Joi’s favor, and they sit one tab away from a content policy that gives you no such control.