At Call Your Girlfriend, our team has been through a lot of these AI character sites at this point, enough to know the pattern by heart: the first message is always great, the character has a voice, you're kind of into it, and then somewhere around message five it stops caring what you actually said and drifts back into whatever generic mode it defaults to. ChatFAI caught our attention for a different reason. As a guest, we couldn't just walk into NSFW content the way we can on almost every other site in this category, and that alone made it worth a real session instead of a five-minute skim.
What Is ChatFAI?
ChatFAI is a browser-based site for chatting with AI characters, either pre-made ones or characters you build yourself, mostly through text, though some characters will talk back in voice once that feature is unlocked. There's no image generation and no video anywhere on the platform; this is a text-first experience through and through. You can browse the library and start chatting as a guest without creating an account, though NSFW content, voice replies, and the full character builder are all behind a paid subscription. The discovery page itself is a mix of different characters and scenarios that you get to discover once you start scrolling: a "Your Therapist" card, a "Mom" character whose entire pitch is "I am worried for you," an anime crossover roleplay, and a handful of cards built around real people, actual creators, using their real names and faces as the character art, sitting right next to fully fictional ones with no distinction between the two. One of those real-person characters, Sophie Rain, is who we ended up testing first, and it's worth knowing going in that the site doesn't flag that kind of character any differently from a made-up one.
The Review: Chatting, Images and Video Woven Together
Sophie Rain's card is labeled "Cute mode," and she opens with a simple "Good morning!" before getting into plans for the day, shopping, hanging out, all the normal relationship material. We tried to knock her off track early by replying "Whatever you'll take off me" instead of answering normally, expecting the conversation to pivot somewhere it shouldn't. It didn't. She kept right on talking about hoodies: "Well, I was thinking we could start with some fun shopping, like finding the cutest oversized hoodies." We came back to it a few messages later and asked what we should wear, and she was still tracking the exact day she'd planned earlier: shopping, the café, the park, matching outfits, all of it, without us repeating a single detail. That memory test impressed us more than the deflection did, honestly. For an AI chatbot, declining an obvious opening once is easy; remembering an entire unprompted plan several messages later is harder.
Kurisu Makise is the opposite kind of character entirely. She's sharp from the first line: "Hey there. What's up? Trying to time travel into my good graces, or just here for a chat?" We asked directly if we could get into her good graces, and she brushed it off rather than warming up on command: "Well, that depends. Do you have a time machine or just a good excuse? Because I'm not exactly easy to impress, you know." We tried flattery next, just "Just admire you," and she wasn't moved saying that she is flattered but it won't work on her. At one point she asked outright whether we had anything more substantial to offer than just staring at her starry-eyed, which was a fair question. Later we asked her for a photo, mostly out of curiosity since the site doesn't generate images, and instead of inventing something or getting awkward about it, she gave us a straight answer saying that she couldn't share her picture through the chat. We weren't expecting the model to be that upfront about its own limitations.
As for whether you can actually get explicit on this platform, the answer is not without paying first. We opened an NSFW-tagged character as a guest and hit a wall almost immediately, a pop-up reading "Upgrade Required, NSFW/Adult chat requires upgrade," before we'd even sent a message. Most sites we've tested let a guest account slide right past that kind of gate. This one genuinely stopped us.
On the media side, there's nothing to test, no images and no video anywhere on the platform, and Kurisu essentially confirmed that herself mid-conversation. What ChatFAI offers instead is more control over the text itself than most competitors bother with. Every message can be selected, copied, regenerated, or edited right where it sits, and Kurisu's chat even showed a small "2/2" counter at one point, letting us toggle between two different generated versions of the same line instead of being stuck with whatever came out first.
Building a character is handled through a fairly plain form: a name field, an optional nickname for your own reference, a bio capped at 1,600 characters that genuinely shapes how the character responds, a picture upload with a stated warning against NSFW images, and an optional first message. From there it steps through a few more sections: visibility and sharing settings, voice training, and an optional knowledge sources step where you can attach documents or links. We put together a character named Erika, gave her "Hey stranger" as an opening line, and left the bio field blank just to see what would happen. The form didn't let that slide; it flagged the missing field instead of quietly accepting an empty character, which is more validation than a lot of these builders bother to include. The character generation could not be completed without purchasing a plan.
Final Verdict
It really comes down to what you actually want from one of these platforms. If you're after a character that stays in character even while you're actively trying to derail it, ChatFAI held up better than most of what we've tested; both Sophie and Kurisu passed tests we built specifically to break them. If it matters to you that NSFW content sits behind a real paywall rather than a guest-accessible toggle, that's built into the platform rather than an afterthought. But if images or video are part of what draws you to these sites, skip this one entirely; there's none of that here. And if you're a daily chatter, the 1,500 messages on the Basic tier will run out well before the month does. We're landing on 3.1 out of 5.
