To create an AI talking avatar for a training course, start by deciding what you want to end up with. A video: you write a script, the tool has an avatar say it, and you get a file or a link. Or a conversational avatar: it presents the course and answers learners' questions. You can try the video route for free.
Video or conversation: which AI avatar suits your training?
An avatar video delivers the same script to every learner. It works well for presenting a concept, a procedure or an instruction.
With a conversational AI avatar, each learner asks their own questions, and the avatar quizzes them in return. It doesn't produce a file: everything happens in an online session.
The choice comes down to one question. Do your learners need to be able to ask "why?", or to answer out loud? If not, a video is enough.
How to create an AI talking avatar video in 5 steps
For a talk of a few minutes, an avatar video holds up. A study by Torbjørn Netland and colleagues, published in 2025, recruited 447 participants online in the United States. They watched short videos from a management course. Some were filmed, others generated by AI, with an avatar created from a photo of the instructor. End-of-session quiz results were comparable. Participants did rate the learning experience slightly lower with the generated videos, by about 4%.
- Write the script, one concept per video. Start from the module's objective and outline: our step-by-step tutorial for creating a training course helps you pin them down. Aim for short sentences written to be spoken, and read them out loud.
- Choose the avatar. Pick one from the tool's library, or create an AI avatar that looks like you. According to HeyGen's avatar creation page, you then need a short video of yourself and a consent form. The video should be well lit, against a plain background.
- Set the voice and language. Listen to every acronym, technical term and proper name. If one is mispronounced, spell it the way it sounds.
- Generate, then watch the whole video. Every attempt eats into your free allowance: note all your fixes so you only regenerate once.
- Publish. Share link, video file or SCORM package for your learning management system (LMS): your plan decides which (see below).
Factor in production time too. The videos in the study ran from 2 min 29 s to 3 min 23 s. A generated video took 3 to 4 hours on average, review included, and never more than 8. A filmed video took about 16 to 28 hours, including scripting, recording and production.
An avatar that answers learners: role, sources, tests
Creating an AI character that talks with your learners doesn't involve a fixed script. You define its role and its sources, then you put it to the test.
- Its role: what it presents, the questions it asks, what it says when someone goes off topic.
- Its sources: your documents, with an instruction to stick to them.
- Tests: ask it your learners' real questions, plus two or three off-topic ones, and check each answer against your documents.
- Follow-up: after launch, read through the questions learners asked the avatar, if the tool keeps them. They show what the course explains poorly.
As for tools, Synthesia's interactive avatar can listen, speak and respond, and it runs through an API. It plugs into your own conversational agent, with the language model and knowledge base of your choice. For now, its integration module works with a single developer tool, in Python. A launch offer gives 500 free minutes.
Since October 1, 2026, according to Synthesia's help center, its self-serve plans also include 60 minutes of interactive avatar time, granted once at sign-up. Without a developer, check what this mode allows, or start from a platform that already combines the avatar and the course.
What do the free plans let you do?
A free AI talking avatar is enough for a first test. To find out whether the video can work in a module, look at four points.
Synthesia, Basic plan.
- Volume: 10 minutes of video per month, plus 500 credits to spend as you like.
- Avatars: 9 AI avatars.
- Watermark: videos carry Synthesia's watermark.
- Distribution: by share link, with no download and no SCORM export.
HeyGen, Free plan.
- Volume: 3 videos per month, one minute long at most.
- Avatars: more than 500 library avatars, and 1 custom video avatar, according to the pricing page.
- Watermark: removing it requires a paid plan.
- Distribution: no SCORM export, which starts with the Business plan.
Free plans and features checked on October 7, 2026, on the Synthesia and HeyGen websites. Check the current terms before you start.
Your first test: one concept, one minute, three learners
- Pick a concept your learners struggle with, and write a one-minute script.
- Generate the video on a free plan, then show it to three learners.
- Write down their questions. If they have a lot of them, what you need is a conversational avatar.
Complement combines the avatar and the course: its AI authoring tool generates the slides and the script from your documents. The avatar presents the content and answers questions within the limits of your sources. It also quizzes the learner and comments on their answers out loud. You get a session, not a video file. It can be shared by link or, priced on request, in an LMS that supports LTI 1.3. The avatars speak French or English.
To try a conversational avatar on your own documents: 7-day free trial, no credit card required, with up to 5 learners.