Guest Post
If you've sat in on AI product meetings this year, you've probably heard someone ask for an "AI video call" on the roadmap. The phrase sounds precise. It isn't.
In practice, those four words cover a live conversation with a visible character, a generated clip that talks after a prompt, an optional camera feed, and a meeting assistant layered on top of a human call. These are different products. They cost different amounts, make different promises to users, and fail in public in different ways.
That matters for founders and marketers, including teams across Southeast Asia who are being asked to add AI video to campaigns, apps, or community products. Brief the wrong category and you end up promising users a conversation, then handing them a file they can't answer.
The naming problem: "AI video call" means four different things
Search results and sales decks still lump the category together. So stop asking whether a tool "has AI video" and start asking which of the four jobs it actually does. A product can offer one without the others, which is how a demo can look impressive and still miss what the user thought they were buying.
Live avatar output
This is the version most people picture, and the one they're least likely to have actually used. A fictional character answers you while the conversation stays open. You see a face and hear a voice as the reply happens, instead of waiting for a render to finish. You speak or type, they answer, and you keep going.
Optional camera input
Camera input is its own feature, even though "video call" makes it sound mandatory. You should be able to see and hear the character with your own camera off. Granting microphone permission shouldn't switch the camera on, and sharing, where it exists, should be an explicit choice you can turn off without ending the session.
Generated video clip
Most marketing teams already understand this category, because it looks a lot like content production. You write a prompt or upload a still, a tool renders a talking clip, and you get a file when it's done. That file can be useful. It can also be mistaken for a call, simply because a face is moving on screen. But it isn't one: once generation has started, the viewer can't interrupt it or take the scene anywhere else.
AI on your own calls
The fourth category sits closest to workplace software. Meeting assistants, translators, and video-agent platforms add AI to conversations people are already having. You might get captions, notes, or a support voice. What you won't get is a persistent fictional character whose job is to stay in a scene with you.
Why the distinction matters before you write the brief
Category confusion is an operations problem. A team that thinks it's buying a conversation will staff and measure the work as though users can talk back. A team buying generated media should be lining up editors, reviewers, and a publishing calendar.
The failures look different, too. A weak generated clip just reads as mediocre content. A weak live conversation feels like a broken relationship: lag, a character you can't interrupt, or memory that pretends to be perfect.
Before you pick a vendor, decide whether you're asking people to watch something or to stay in a scene and answer. If you can't say which job you're hiring the AI to do, you aren't ready to buy.
What a live character experience actually feels like
A convincing live session is a conversation loop, not a finished clip. Obvious enough, until you sit through a demo, where a polished ten-second sample could pass for either one.
The test is simple. Can the user change what happens next while the character is still talking? If not, you're looking at media. If so, you're looking at a conversational interface that happens to have a face.
In a live character call, you start by speaking or typing. The reply comes back as a face and a voice, with expression as the character talks. You can cut in mid-sentence, add a thought, or move the scene along, and camera sharing can stay optional.
Memory deserves the same care. A character that "remembers everything" is a marketing line, not a specification. Recall is often selective, chat history may not carry over into a call automatically, and details that need to travel usually have to be pinned or restated.
Prelulu is one consumer example of this first category, as opposed to clip generation. It's built around original fictional characters you can text and then meet on a live voice or video call. The company behind it, TheInfluence, Inc. in San Mateo, trains its own realtime avatar model rather than treating the face as a finished video exported after a prompt.
Inside its AI video call app, the character looks at you and answers in their own voice. You can speak or type, interrupt while they're talking, and keep your camera off. As of 9 November 2026, more than 150 characters in the public catalog can be started as a video call.
The website, iOS app, and Android app all support the same basic loop, and the site comes in 27 interface languages. That's the minimum "live" has to include before it earns the word: a visible character, a voice, an open turn, and a way to cut in.
Live conversation vs generated clip: a decision checklist
Most teams don't need one winner for the whole company. They need the right format for a specific job: a file the user watches, or a turn the user can answer.
When a generated clip is the better tool
Pick a generated clip when the audience is supposed to watch, not respond. Think trailers, explainers, social ads, launch films, and any asset that has to be reviewed, edited, and published on a schedule.
Generated video also makes sense when legal and brand review need to see the exact frames before anything goes live. You can't fully approve a live conversation in advance, because the next line doesn't exist until the user speaks. If you're measuring watch time or click-through on a piece of media, treat the AI as a production assistant and hold it to the same quality bar as any other video.
When a live conversation is the better tool
Pick a live character conversation when the value lies in continuity. The user should be able to stay with a character, steer a scene, ask a follow-up, or hear a reply that only exists because of what they just said. That's closer to interactive entertainment, roleplay, dialogue-heavy learning, or any branded experience built around presence rather than playback.
Here, what you want is interactive AI video stories in the conversational sense: fiction the user moves forward by talking back, where any video is the live character answering, not a story rendered from a prompt. That's a different product from an "AI video story generator," even though the search terms overlap. The clearest consumer signal comes from companion and romance fiction, where people looking for an AI boyfriend chat now expect the same character to text, hold a storyline, and then pick up a live call.
Common mistakes in the brief
The most common mistake is pitching with celebrity or copyrighted-character language. "Talk to your favourite idol" might sound like an easy acquisition hook, but it creates legal risk and teaches users to expect a likeness the product shouldn't offer. Original characters aren't a watered-down version of that idea. They're the version that can actually ship.
The second mistake is selling cinematic intimacy while hiding the limits. If a call can only run so long, if video burns more credits than voice, or if memory is selective, say so. Users forgive limits they can see. What they won't forgive is a trailer that falls apart in the first live minute.
The third is measuring a conversation as if it were a campaign asset. A live session should be judged on whether people can start easily, stay in character, interrupt naturally, and come back without the story resetting for no reason.
UX details marketers should inspect before they promise anything
Inspect the demo as a conversation product, not a render farm.
First, test interruption. Talk over the character. If the system can't yield, users will feel stuck inside a monologue, which is the opposite of a call.
Second, test input flexibility. Some users want to talk. Others want to type, especially in public or on a commute. A product that only works when you speak out loud shrinks its own audience.
Third, check camera behaviour on a phone with the permission prompts in view. Make sure seeing the character doesn't require showing your face, and that turning sharing off doesn't end the call.
Fourth, ask how memory works across text and calls. If a user spends ten minutes setting up a scene in chat, does any of that context reach the call? If not, the product needs a pinning step or a spoken recap, and your onboarding copy should say so.
Fifth, check language and device reality, not just the feature list. A character that answers in the language you speak beats one that only looks polished in a desktop demo. And browser access with no install still matters when someone should be able to try it the same day they hear about it.
How these products are actually tried and paid for
Live avatar sessions tend to be priced like usage, not like a stock video library. A pilot should check whether a new user can get started, whether a short first session is possible without a purchase, and whether the balance makes sense before someone hits call.
Under Prelulu's public web pricing as of November 2026, the product is free to start. Text chat comes with a daily message allowance and no card required, and welcome credits on signup can cover a short call. Live voice and video then draw credits from a single wallet, with video using more credits per minute than voice.
Web membership starts at $9.99 a month, or $5.99 a month billed yearly, and the entry plan includes 7,200 credits. That works out to about 60 minutes of video on characters that offer it, or about 180 minutes of voice, and you can mix the two. Each call can run up to 30 minutes.
Prices in the iPhone and Android apps are a little higher because the stores take a cut. Otherwise, the plans and credits are the same.
If pilot users can't tell what a minute costs before they start, they'll see the product as unpredictable and leave. Let people begin in text, make the first live minute easy to reach, show the balance before the session starts, and don't spring a camera request or a hidden credit drop on anyone.
What this means for teams in Southeast Asia, and everywhere else
Southeast Asian teams are often asked to move faster than their tooling allows. Mobile-first usage, shared-phone households, and multilingual audiences aren't edge cases in Jakarta, Bangkok, Ho Chi Minh City, Kuala Lumpur, or Singapore. They're the everyday conditions in which a character product gets tried.
So browser access, iOS and Android apps, and broad language coverage aren't extras. They decide whether the experience exists at all for the people you say you serve. A site available in 27 interface languages, including Bahasa Indonesia, Bahasa Melayu, Vietnamese, and Thai, tells you more than any claim about being global.
People in the region already know the grammar of idol livestreams, drama characters, webtoon tropes, and long-running chat roleplay. Use it with original characters, not copyrighted figures or real celebrities. Original characters can be reused, improved, and owned. A lookalike strategy can't.
A live call also needs a steadier connection than a pre-rendered clip. If your users hop between wifi, mobile data, and crowded venues, tell them what the call needs and give them a typed fallback.
A two-week way to evaluate the category without overcommitting
In week one, map the job in a single sentence: is the user watching or answering? List where a generated clip would be safer, like ads or anything that needs frame-by-frame approval, and where only a live reply will do, like follow-up questions or a scene the user wants to stay inside.
Then try the live path the way a new user would. Open it in a browser on a phone, keep the camera off, speak once, type once, and interrupt mid-sentence. Come back the next day and see whether anything you cared about survived.
In week two, bring in brand, legal, support, and whoever owns the budget. Have them watch a live session, not a curated recording. Ask whether the characters are clearly fictional, what happens when memory fails or credits run out, and what a realistic month would cost at published rates, not at a demo's best-case minute count.
If the live path still fits, write the public language last, using words that match the product: live call, original character, optional camera, credits, selective memory. The quickest way to waste good technology is to advertise it as something else.
The point is not more video. It is a clearer promise.
AI video isn't rare anymore. What's still rare is a visible character who can answer you in the moment, stop when you interrupt, and carry on a story you're actually part of. Generated clips are still the right tool whenever you need a finished piece of media. Live avatars are the right tool when the conversation itself is the product.
Marketers and founders don't need to become avatar engineers to make that call. They just need to stop treating "AI video" as a single feature. Separate the file from the session, the character from the meeting assistant, and original fiction from a lookalike you can't defend.
Do that before the campaign deck is locked, and the technology gets a fairer shot at being judged on what it actually does. Users can tell almost right away whether someone is answering them or a clip has simply started to play.
Frequently Asked Questions
Can you FaceTime an AI?
Not through Apple's FaceTime itself, but the experience people mean by that phrase does exist. A live AI video call opens a session in a browser or app where an animated character answers in real time with a face and a voice. Your own camera is usually optional, so "FaceTime with an AI" normally means you can see the character, not that the character has to see you.
What is the difference between an AI video chat and an AI video generator?
An AI video chat is a session: the character responds to what you just said, and you can interrupt, ask a follow-up, or change direction. An AI video generator produces a file from a prompt, which you watch once it has rendered. If the viewer cannot talk back, it is generated media, not a chat, however lifelike the face looks.
Are there free AI video call apps, and do they need a sign-up?
Several consumer apps let you start for free, usually with text chat first and a small allowance of credits for a short live call. Live video costs far more to run than text, so free minutes tend to be limited, and most services ask for an account before the first call even when browsing characters is open to everyone. Check what a minute costs before you plan a pilot around it.
Can an AI character remember previous conversations?
Partly. Most products keep recent context and let users pin or restate the details that matter, but recall is selective rather than total. The question worth asking a vendor is whether chat history carries into a voice or video call, and what happens to a storyline after a day or a week away.
What is AI video roleplay?
It is interactive fiction delivered as a live call: you speak or type, and a character answers in voice and on screen while staying inside a scene you are shaping together. Unlike a rendered story clip, the plot only moves because of what you say, which is why it suits original characters and long-running storylines better than one-off promotional videos.
Why are AI boyfriend and companion apps adding live video calls?
Companion and romance fiction is where presence matters most. Users who have texted the same character for weeks want to hear and see them, and a real-time call is the natural next step. Prelulu, for instance, lets the same original character move from chat to a live voice or video call. For marketers outside that category, the lesson transfers: when the relationship with a character is the product, a live reply beats a polished clip.