How to make an AI avatar video in Synthesia (2026)

The fastest reliable workflow starts from an approved script and a short representative scene. Build and review that scene before converting an entire course or deck.

What you need before opening Synthesia

• Audience and outcome: one sentence describing what the viewer should know or do.

• Approved script: facts, claims, names and pronunciations checked by the owner.

• Brand assets: logo, colors, fonts, product captures and any licensed media.

• Presenter choice: a stock, personal or custom avatar with the required consent and usage rights.

• Distribution format: landscape, portrait or square, plus captions and localization needs.

Step 1: choose the right starting point

Synthesia’s current video-creation documentation lists prompt, template, script, file, PowerPoint and blank-canvas starting points. Use a script when wording is already approved; use a template for repeatable layout; import a deck or document only when its structure deserves to survive.

A prompt-generated draft is useful for options, not factual approval. Compare it with the source and remove invented claims before assigning a presenter.

Step 2: set the aspect ratio and scene system

Choose the output format before laying out scenes. Synthesia’s editor guide recommends setting the aspect ratio before design and describes a scene list, canvas, inspector and script box.

Use one scene for one idea. Keep a stable grid for presenter position, headline, evidence and supporting visual so updates do not require redesigning every slide.

Step 3: choose an avatar and document consent

Synthesia’s avatar page describes stock, customizable and personal avatars. A personal avatar workflow requires a consent recording; enterprise controls and allowances vary by plan.

Match the presenter to the audience and context. Avoid a synthetic spokesperson for testimony, news, sensitive advice or any situation where the viewer could reasonably infer a real endorsement.

Step 4: write for spoken delivery

• Lead with the answer. State the action or conclusion in the first sentence.

• Use short clauses. Split dense policy or product language into separate spoken beats.

• Mark pronunciation. Test names, acronyms, URLs, numbers and domain terms.

• Show evidence. Put the relevant interface, chart or source on screen while the presenter explains it.

• End with one action. Avoid several competing calls to action.

Step 5: assign voice and timing

Paste the script scene by scene and select the voice and language in the script box. Preview difficult lines before styling the whole video. Adjust the copy before forcing unnatural speed or pauses.

For localization, translate meaning and on-screen layout together. Names, dates, currency, captions and text expansion require a human language review even when the voice sounds fluent.

Step 6: add evidence and B-roll

Use product captures, diagrams and licensed footage to prove what the narration says. Synthesia also documents avatar and space B-roll through supported generative models in the editor.

Generated B-roll should have a written job: demonstrate a concept, bridge a cut or maintain attention. If the script needs an original cinematic insert, compare it with a dedicated AI video generator and keep the shot that survives factual, continuity and rights review.

Step 7: preview the complete experience

• Meaning: script, image and presenter communicate the same fact.

• Voice: pronunciation, pace, emphasis and pauses sound intentional.

• Visuals: no cropped UI, tiny text, accidental brands or unsupported claims.

• Transitions: cuts and scene changes do not interrupt a sentence.

• Accessibility: captions match the final audio; color contrast and reading order work.

• Disclosure: label synthetic presenters where context, policy or law requires it.

Step 8: generate, publish and measure

Generate only after the representative scene passes review. Synthesia’s current AI video feature page describes avatar-led videos plus generated assets, and its pricing page is the source of truth for current credits, downloads, collaboration and plan access.

Measure completion rate, the first major drop, comprehension or task completion, edit time, correction count and cost per approved minute. For training, also measure whether viewers can perform the task after watching.

When Synthesia is a good fit

• Strong fit: repeatable training, onboarding, internal communication, product explanation and localized presenter videos.

• Weak fit: documentary proof, authentic customer testimony, live events, unscripted emotion or a story that depends on physical performance.

• Separate job: cinematic generative clips and complex product footage may need a dedicated model or conventional production.