How to make a talking presenter video without getting on camera

This page may contain affiliate links from which Blur Studio earns commissions.

Not everyone wants to be on camera, and not everyone has the time to reshoot a presenter video every time a script changes. An AI avatar solves both problems at once: build the presenter once, then generate as many versions of the script as you need without booking a studio or a person again.

How do ElevenLabs' Avatars actually work?

ElevenLabs added Avatars to its ElevenCreative platform in June 2026. An avatar is assembled from source images you provide, and it comes with a set of styles, meaning the same "person" can appear in different scenes and outfits. Each avatar carries a default voice that you can swap for any voice in ElevenLabs' library. You write a script, the tool generates lip-synced speech from it, and the talking-head video itself renders through a third-party engine, Creatify Aurora, working underneath ElevenLabs' own audio generation.

The number worth knowing before you script anything: speech generation is billed at 1 credit per character, and each single generation is capped at 5,000 characters and 3 minutes. A typical two-minute presenter script runs well inside that limit, but anything closer to a full webinar intro needs to be planned and generated in segments, which affects how you edit the final video together.

What I'd actually tell you if this is your situation

For explainer videos, course modules, internal training updates, or any context where the content matters more than the performance, an AI avatar is a genuinely good trade: consistent presence, instant re-recording when the script changes, no scheduling a person every time. I'd steer you away from it for anything where the audience's trust depends on recognizing a real, familiar person, a CEO's quarterly update to a team that already knows their face and voice well, or a sales context where "is this actually a real person" becomes the first objection instead of the message itself. In both of those cases, the uncanny-valley risk costs you more credibility than the convenience saves you time.

If the presenter is based on a real person's face, read the consent rules for talking presenter videos first. Test the avatar on a short clip (30 to 60 seconds) before scripting a full video. Lip-sync quality and the avatar's naturalness vary enough across different source images and voices that you want to know it works for your specific setup before you commit a longer script to it.

Step by step: making a talking presenter video

1. Write the full script first, the way you'd write something to be spoken aloud, not read.
2. Choose source images for your avatar that are clear, well-lit and front-facing; avatar quality starts with the input images, not just the generation.
3. Pick a voice from ElevenLabs' library, or keep the avatar's default if it already fits your tone.
4. Break scripts over roughly two minutes into segments that each fit the 5,000-character, 3-minute generation cap.
5. Generate a short test clip first and watch it on the device your actual audience will use (a phone screen hides flaws that a monitor reveals, and vice versa).
6. Once you're happy with quality, generate the full script and stitch segments together if you had to split it.

Avatar video generation needs at least a Starter plan on ElevenLabs; the Free plan covers image generation only.

How does this compare to HeyGen and Synthesia?

HeyGen and Synthesia are both established, purpose-built AI avatar and presenter video platforms that predate ElevenLabs' entry into this space by some margin, and they generally offer deeper avatar customization options (clothing, backgrounds, more granular control over gesture and expression) since that's the entire product, not one feature of a broader voice platform. ElevenLabs' advantage is that its underlying voice quality, the thing it's genuinely best at, carries over directly into the avatar's speech, so if voice naturalness matters more to you than avatar visual polish, it's worth testing against either competitor directly rather than assuming the more established name automatically wins.

On price, HeyGen publishes a Free plan at $0, Creator at $29 a month, Pro at $49 a month and Business at $149 a month, with Enterprise custom-quoted above that. ElevenLabs' avatar feature rides on top of its own Starter plan at $6 a month, which is a meaningfully lower entry point if you're testing the idea before committing to a dedicated avatar subscription. That gap narrows once you factor in what each plan actually includes: HeyGen's paid tiers are built around avatar minutes and seats specifically, while ElevenLabs' credits cover your whole account's usage across every feature, not just avatars, so a heavy avatar workload can burn through ElevenLabs credits faster than the headline price suggests. If avatar video is the only thing you need from either tool, run the actual math on your expected monthly output (minutes of finished video, not just the subscription price) before assuming the cheaper entry tier is the cheaper real cost.

A worked comparison: say you need 10 two-minute course-intro videos a month. At roughly 1 credit per character and a typical 300-word (about 1,800-character) script per video, that's 18,000 credits a month, comfortably inside Starter's monthly allowance with room to spare for a few retries. The same 10 videos on HeyGen's Creator plan at $29 a month would need to fit inside that plan's included avatar minutes, which is a harder ceiling to hit since it's minutes of finished video, not characters of script, and retries or longer takes eat into that total faster. For a handful of short videos a month, ElevenLabs is often the cheaper way to prototype the idea; for a heavier, ongoing video production schedule where avatar polish matters more than cost, HeyGen or Synthesia's dedicated feature set starts to earn its higher price.

When should you test carefully before committing?

ElevenLabs' own public description of Avatars is still fairly brief next to how long HeyGen and Synthesia have been refining this exact feature. Test quality on your specific use case rather than assuming parity with more established tools, especially if you're planning a longer or more important video around it.

A worked example: a 10-module course

Say you're building an online course with 10 modules, each needing a 90-second presenter intro. Filming that yourself means 10 separate recording sessions, or one long session you later cut into 10 pieces, and any later edit to module 3's script means reshooting just that section and hoping the lighting and energy match the rest. With an avatar, you build the presenter once, write all 10 scripts, and generate each intro separately since each sits comfortably under the 5,000-character cap on its own. If you update module 7's content three months later, you regenerate just that one clip with the new script and the presenter looks identical to the other nine, no reshoot, no studio booking, no trying to match your energy from three months ago. The real saving isn't the first batch of 10, it's every revision after that, which is where a filmed course actually costs you time. If you already have a filmed training video, you can reuse it with a brand mascot instead.

Writing a script for an avatar presenter, not a human one

A script written to be read by a human presenter and a script written for an avatar should look slightly different. Avoid long, unbroken sentences; an avatar's pacing tends to read more naturally on shorter clauses with a clear pause point between them. Write out numbers and abbreviations the way you want them spoken (“twenty twenty-six” rather than “2026” if that's the pronunciation you want), since the avatar reads the text as written, not as a human would interpret it in context. Avoid stage directions or parenthetical asides in the script itself; put a gesture or emphasis instruction in the voice settings or in a separate note, not inline in the text that gets spoken aloud. Finally, read the script out loud yourself before generating, even if you'll never be on camera: if it's awkward in your own mouth, it will sound awkward from the avatar too, since the underlying speech model is matching your words, not improving on them.

Spotting and fixing lip-sync and quality issues

The most common quality problem isn't the voice, it's the source images. Flat, low-contrast lighting or a face at a three-quarter angle instead of straight-on produces a noticeably less convincing lip-sync than a clear, evenly lit, front-facing photo. If a generated clip looks slightly "off" around the mouth during fast speech, try slowing the pacing in the voice settings rather than assuming the avatar itself is the problem; fast consonant-heavy phrases are where lip-sync most commonly breaks down. If a specific word or name consistently renders oddly, that's usually a pronunciation issue with the underlying voice model rather than the avatar layer, and it's worth testing the same line as plain text-to-speech first to isolate which part of the pipeline needs the fix. Keep your test clips short (30 to 60 seconds) specifically so a bad take costs you a minute of waiting, not ten.

How I'd approach this

A single presenter video is a tool you can run yourself in an afternoon. Where I get involved is when someone wants this generating automatically at scale, personalized presenter videos for every lead in a CRM, or a course platform that regenerates the presenter's script every time the curriculum updates. That's a real build on ElevenLabs' API, wired into whatever system is actually producing the scripts, not a one-off video someone makes by hand.