AI Avatars FAQ: Library, Custom Clones, Realism and Limits
Everything about Klip Kanvas AI avatars — the 50+ library, custom cloning, consent rules, delivery control, and where avatars still look obviously fake.
Casting is the biggest single decision in an AI UGC ad, and almost every avatar question we get falls into seven areas: the library, cloning, realism, delivery, consistency, rights and plan access. Here is the honest version of each, limits included.
01
Avatar Library
1.How many AI avatars can I choose from?
The library ships with 50+ avatars and grows every month. They span age brackets from early twenties to sixties, a wide range of ethnicities, and both casual-creator and expert-presenter looks. Each avatar is pre-shot in one or more environments — car interior, bathroom mirror, kitchen counter, home desk, outdoor street — so the scene reads as a real phone recording rather than a studio green screen. In practice most accounts settle on a working set of 4–6 avatars they rotate through, and keep the rest for testing new angles or new markets.
#Avatar Library
2.What settings and scenes do the avatars appear in?
Every avatar is tied to a set of real environments rather than a blank backdrop, because the background does a large share of the work in making UGC read as authentic. Common ones are car interior, bathroom mirror, kitchen, bedroom, home-office desk, gym and outdoor street. Match the setting to the moment of use: a supplement review lands better in a kitchen or a car, a skincare routine in a bathroom, a B2B tool at a desk. A mismatch — someone reviewing a kitchen gadget in a gym — breaks the illusion in the first second, and viewers register it before they hear a word.
#Avatar Library
3.How do I pick the right avatar for my product?
Cast for your buyer, not for your taste. The highest-performing avatar is usually someone who looks like your customer three to five years into owning the product, not a model. Start by matching the age band and gender presentation to your top-selling segment, then match the environment to where the product is genuinely used. Then test: our standard starting batch is 6 creatives — 3 hooks × 2 avatars on a single body script. If one avatar wins across all three hooks, that is a casting signal, not a copy signal, and it is worth more than any preview you can watch in the editor.
#Avatar Library
4.How often are new avatars added?
New avatars land monthly, and existing ones are occasionally re-rendered as the underlying model improves. That cadence matters more than it sounds. An avatar that has been running in your account for eight weeks at frequency 2.5–3.5 is already contributing to fatigue, and swapping the face while keeping a proven script is the cheapest refresh available to you. Keep a simple log of which avatars you have burned through on which audience, so you are not recycling a face your retargeting pool has already seen twenty times.
#Avatar Library#Consistency
5.Can I filter the library by age, gender or look?
Yes. The library filters by age band, gender presentation, apparent ethnicity, energy level and environment, so you can go from 50+ options to a shortlist of five in a few clicks. Save that shortlist per brand — agencies running six clients usually keep a fixed cast per client so creative stays recognisable across months. One caveat worth stating plainly: filters describe how an avatar looks, not how it performs in your vertical. Appearance gets you a candidate list; only a live test tells you which face your audience actually responds to.
#Avatar Library
02
Custom Avatars
6.Can I create a custom avatar of myself or my founder?
Yes, on higher-tier plans. You record a short reference video reading a consent script on camera, upload it, and the platform builds a private avatar only your workspace can use. Founder avatars earn their keep in categories where the person is the trust signal — coaching, agencies, supplements with a formulator story, local services. In high-volume commodity categories a stock avatar usually performs the same, so do not gate your first campaign on getting a clone approved. Check /pricing for which tier includes cloning today.
#Custom Avatars#Plans & Access
7.What do I need to record for a custom avatar?
A stable, well-lit reference recording of two to three minutes: neutral background, phone at eye level, natural speaking pace, and the consent lines read aloud so the identity check has something to verify against. Shoot in the framing you want the avatar to appear in — a waist-up recording cannot be turned into a full-body avatar later. The three things that ruin a clone are backlighting, background noise and reading in a monotone. If the source performance is flat, every ad you generate from it will be flat too; the model copies energy as faithfully as it copies a face.
#Custom Avatars
8.How long does a custom avatar take to build?
Budget one to two business days from upload to a usable avatar, including the consent review step. That is deliberately slower than stock avatars, which are available instantly, because a human check on likeness consent sits in the middle of the pipeline and we will not automate it away. Plan around it: if a launch is two days out, build the campaign on stock avatars and swap the clone in on the next iteration cycle. Once the avatar exists, generating with it is exactly as fast as any other — 3–5 minutes for a typical 30-second ad.
#Custom Avatars
9.Can other Klip Kanvas users see or use my custom avatar?
No. A cloned avatar is scoped to the workspace that created it and is never added to the public library. Team members on your seats can generate with it; nobody outside can see that it exists. If you are an agency cloning a client's founder, create it inside that client's own workspace rather than your shared one, so access ends cleanly when the contract does. We delete custom avatars on request, and deletion removes them from all future generations — videos you already rendered and downloaded stay yours.
#Custom Avatars#Rights & Consent
03
Realism & Limits
10.How realistic are the avatars in an actual feed?
In the context they are built for — a 9:16 feed, watched on mute, at arm's length, for about two seconds before a scroll decision — current-generation avatars pass for filmed creators with most viewers. Micro-expressions, hand gestures, accurate lip-sync and environment-matched lighting are all present. Context is the whole answer, though. Played frame by frame on a laptop, a trained eye finds the tells within seconds. Since almost all paid social is consumed the first way, the useful question is not “is this indistinguishable” but “does it hold attention for three seconds”.
#Realism & Limits
11.Where do AI avatars still look obviously artificial?
Four places, consistently: hands doing fine manipulation, teeth during wide smiles, hair moving against a busy background, and any moment where the avatar physically interacts with an object. Long unbroken takes also expose it — past roughly 8–10 seconds on a single shot, small repetitions in gesture start to register even on a phone. The workaround is editorial rather than technical: cut to B-roll every 3–5 seconds, keep talking-head segments short, and never ask an avatar to demonstrate something with its hands. If a shot depends on manipulation, film that one shot yourself.
#Realism & Limits
12.Can an avatar hold and demonstrate my actual product?
Not convincingly, and this is the biggest limitation to plan around. An avatar can gesture toward a product and can appear in the same frame as one, but a genuine demo — unscrewing a lid, applying cream, assembling parts, pouring a scoop — is where AI generation breaks down and viewers notice. The fix is a hybrid edit: the avatar delivers the hook and the claim, your own uploaded footage carries the demo. In demo-led categories a 30-second ad built from roughly 12 seconds of avatar and 18 seconds of real product footage typically beats a pure-avatar version.
#Realism & Limits
13.Do avatars work for full-body or movement shots?
They are built for waist-up and head-and-shoulders framing, which is what real UGC looks like on a phone anyway. Full-body walking shots, dance, sports movement or anything requiring accurate body mechanics sit outside what the avatar engine does well, and pushing it there produces exactly the uncanny output people screenshot to mock. Apparel is the vertical this hurts most, because fit and drape cannot be faked and are the entire purchase decision. If your creative needs movement, use avatars for the talking beats and real footage for everything else.
#Realism & Limits
04
Delivery & Emotion
14.Can I control the emotion and energy of the delivery?
Yes. Each avatar carries multiple delivery styles — excited, conversational, calm-expert and storytelling — and they are settable per scene rather than once per video. The per-scene control is the point. A strong ad usually opens excited for the two-second hook, drops to conversational for the problem and demo beats, then lifts again for the CTA. Holding a single energy for thirty straight seconds is the most common delivery mistake we see: it reads as an advertisement within about four seconds, which is precisely the window where hook rate is decided.
#Delivery & Emotion
15.My avatar sounds flat and robotic — what actually fixes it?
Nine times out of ten the script is the cause, not the voice model. AI delivery collapses on long clause-heavy sentences and on marketing copy no human would say out loud. Cut every sentence under 15 words, write your contractions in, and read the script aloud before you render — if you stumble, the avatar will too. Then set delivery style per scene instead of leaving one setting across the whole video, and add a deliberate half-second pause after the hook line. Those three changes resolve most flat-delivery complaints without touching the avatar at all.
#Delivery & Emotion
16.Can avatars deliver in different accents?
Yes, across 30+ languages with native-sounding regional voices, and lip-sync is regenerated per language rather than dubbed over the original. Accent choice is a real performance lever: a Mexican Spanish read dropped into a Spain-targeted campaign registers as slightly foreign, and audiences catch it faster than most advertisers expect. Match the voice to the market you are actually buying media in, not to the language generically. Where we cannot help is hyper-local dialect nuance — that still needs a native speaker to review the script before you spend.
#Delivery & Emotion#Consistency
17.Can avatars pause, laugh or react naturally?
Micro-behaviours — beat pauses, a small laugh, a headshake, an eyebrow lift — are supported, and most of them are driven by how you write and punctuate the script. Punctuation does real work: an em dash or an ellipsis produces a longer beat than a comma. Use one deliberate pause right after the hook and one before the CTA, and nothing else; scripts stuffed with reaction cues read as performed rather than casual. Sustained laughter, crying or big physical comedy are outside the range and look wrong, so do not write them in.
#Delivery & Emotion
05
Consistency
18.Can I use the same avatar across all of my ads?
Yes, and there is a real argument for doing it. Avatars are persistent characters, so one face can carry your whole ad account and become a recognisable brand creator — which noticeably helps recall in retargeting, where the same viewer sees you four or five times. The trade-off is fatigue: a single face on a cold audience starts decaying around frequency 2.5–3.5. The setup that works for most accounts is one anchor avatar for retargeting and mid-funnel, plus a rotating cast of two or three for cold prospecting.
#Consistency
19.Does reusing one avatar cause creative fatigue faster?
Somewhat, but far less than reusing one hook. Fatigue is driven mostly by the first two seconds — the same opening line on the same audience is what stops working. Swapping the avatar while keeping a proven body script is a cheap refresh that often buys another two to three weeks of life. When performance slips, work in this order: new hook first, then new avatar, then new body script, then reconsider the offer. Rebuilding a working ad from scratch when only the hook is tired is the most common waste we see in creative teams.
#Consistency
20.How many avatars should I test at once?
Two on your first pass. The standard starting batch is 6 creatives — 3 hooks × 2 avatars on one body script — launched into a single consolidated ad set with broad targeting. Two avatars is enough to separate casting from copy without slicing your budget into unreadable pieces. Read the test after 48–72 hours or roughly 1,000 impressions per variant, whichever comes first. Once you have a winning face, expand the hook count against it rather than throwing four more avatars into the mix at the same time.
#Consistency#Avatar Library
21.Will the avatar look the same when I localise into another language?
Yes. Localisation keeps the same avatar, the same edit and the same pacing, and regenerates the voiceover and lip-sync in the target language. That is what makes market expansion cheap — an English winner becomes a Spanish and Hindi version in minutes instead of a new shoot. The judgement call is whether the same face should carry every market. Sometimes it should, for brand consistency. In markets where casting expectations differ sharply, a locally cast avatar usually lifts hook rate more than the translation quality does.
#Consistency
06
Rights & Consent
22.Whose face is a stock avatar?
Every stock avatar is built from a paid, consenting actor who signed a licence covering commercial advertising use. They are not scraped from the internet and they are not composites assembled from real customers. That licence is what lets you run the output as a paid ad without clearing anything yourself. What it does not do is transfer that person's likeness to you: you hold a licence to use the generated video, not ownership of a human face, so you cannot present an avatar as a specific named individual or as your employee.
#Rights & Consent
23.Can I clone a celebrity, an influencer or a customer?
No. Cloning requires a verified consent recording from the person being cloned and there is no route around that check — not for public figures, not for a customer who said yes over email, not for a former employee. We enforce it because likeness misuse is the fastest way to end up in a legal dispute and to get an ad account restricted. If you want a celebrity association, license it through the normal talent route and film it properly. Uploads that try to bypass the consent step are rejected at review, and the credits are not spent.
#Rights & Consent
24.Do I have to disclose that the person in my ad is AI?
Often yes, and the responsibility sits with the advertiser rather than with us. Meta and TikTok both require disclosure for certain AI-generated content, especially anything that could be read as a real person making a claim, and anything in a regulated vertical such as health, finance or supplements. Rules also differ by region and are still moving month to month. We surface where a disclosure toggle is recommended and we provide the tooling, but we cannot certify that a given ad is compliant in a given market, and our flags are not legal advice.
#Rights & Consent
25.What happens to a founder avatar if that person leaves the company?
Retire it, and request deletion so it cannot be used in new generations. Legally, what you may do depends on what your consent agreement said about duration — most sensible ones limit use to the period of engagement plus a short wind-down window. Practically, running a departed founder as the face of your ads is a reputational risk that outweighs the production cost you saved. The defensive move is to build a second avatar before you need one: if a single person is currently carrying your entire ad account, you have a concentration problem.
#Rights & Consent
07
Plans & Access
26.Which plans include custom avatar cloning?
Cloning sits on higher tiers, alongside 4K export, priority rendering and additional team seats; the free plan and entry tiers use the stock library only. Across this category, entry paid tiers generally sit in the ~$29–79/mo range at time of writing, but check /pricing for our current numbers rather than trusting a figure in an article. If cloning is the only reason you would upgrade, run one full campaign on stock avatars first — for a lot of products the stock library performs identically and the upgrade is not justified yet.
#Plans & Access
27.Do different avatars cost different amounts of credits?
No. Credit cost is driven by video length and output resolution, not by which face you pick, so testing three avatars costs the same as generating the same video three times with one avatar. That is deliberate — casting tests should never be a budget decision. Free accounts get 50 credits with no credit card, which is enough to run a first real two-avatar comparison. On paid plans unused credits roll over for one billing cycle, so a light month does not waste the allowance you were saving for a bigger test batch.
#Plans & Access
28.Can I request an avatar that is not in the library?
You can send requests and library gaps do feed the monthly additions, but there is no guaranteed turnaround and no way to commission a specific look on demand. If a particular look is essential to your campaign — a specific profession, a specific age-and-setting combination — the reliable route is cloning a real person who fits, because you control that end to end. Treat the request channel as input to a roadmap rather than a deliverable, and never build a launch plan around an avatar that does not exist yet.
#Plans & Access#Avatar Library
Ready to put this into practice?
Create your first AI UGC video ad in minutes — no filming, no actors, no editing.
