Source: raw/How_to_Make_an_Employee_Training_Video_With_AI_Full_HeyGen_Tutorial.md (YouTube MzLiwREZxA0, uploaded 2026-10-10) Creator: HeyGen official channel (@heygen_official); presenter not named URL: https://www.youtube.com/watch?v=MzLiwREZxA0 Duration: 7:24 Platform: YouTube Skill level: operator

A HeyGen-made walkthrough that turns a policy PDF into a roughly 3-minute employee training video. The workflow has two stages: Video Agent builds a scene-by-scene plan and first draft from the document, then Edit a copy in AI Studio gives per-scene control for fixing script, voice, media and text. The example is a customer service and returns guide for a fictional retailer, North Star Outfitters. The central rule is to make the PDF the only factual source so the AI organizes policy instead of inventing it.

Key Takeaways

  • Prepare the document first. Write rules explicitly (the example: a 30-calendar-day return window, required proof of purchase, acceptable item condition, refund method, excluded products). Clear, specific rules give Video Agent better source material than vague paragraphs.
  • Example document structure: service standard, return requirements, damaged-product process, manager-escalation situations, practice scenarios, final employee checklist.
  • Video Agent setup choices: a professional talking-head avatar (default voice, replaceable from the voice menu for accent, tone or delivery); a brand system if one exists in HeyGen, otherwise colors in the prompt; chat mode rather than autopilot so you review a plan first; target runtime ~3 minutes; landscape for an internal training portal or LMS.
  • Seedance turned off on purpose. The presenter disables the Seedance feature (caption reads “scene dance”) to get a traditional training structure: presenter, simple graphics, uploaded material and stock footage, not cinematic AI-generated scenes.
  • The prompt names six things: audience, required sections, factual source, visual style, the presenter’s role, and runtime. The key line is “using only the attached PDF as the factual source … Do not invent, remove, or change any policy.”
  • Avatar placement rule: use the avatar for the introduction, section transitions and final recap; use b-roll and motion graphics for policy steps and scenarios.
  • Review the plan before spending credits. Video Agent shows a scene-by-scene plan; check every required section is present and no policy was added, removed or changed. Corrections in chat update the plan without generating the video.
  • Check the credit estimate HeyGen shows before generating, especially with a premium avatar model or a longer video.
  • Five checks on the first draft: policies accurate; one idea per scene; text readable; visuals support the training; presenter appears only where it adds value.
  • Fix scenes, not the whole video. Click edit on one scene and describe the exact change. The presenter says targeted instructions are faster and more predictable than asking the agent to remake the project.
  • AI Studio controls after “Edit a copy in AI Studio”: script (match the PDF exactly); avatar (replace or remove the presenter per scene); voice (voice, accent, tone, speed, voice model, voice engine, with audio preview); media (swap generic stock for uploaded screenshots, videos or pages from the training document); text (section headings, policy callouts); stylized captions (useful for remote training); background music kept quiet; the plus button adds a blank scene or a scene template.
  • The avatar performance cannot be fully previewed before generation, so catch problems in the script and audio preview first: a final policy check against the PDF, then pronunciation, pacing and awkward pauses.
  • Maintenance is the payoff. If the return window changed from 30 to 45 days, you would edit the script instead of scheduling a reshoot. The presenter names onboarding, compliance, safety and software training as the best fits because the information changes regularly.

Try It

  1. Rewrite the source document so every rule is explicit (numbers, conditions, exclusions, escalation triggers).
  2. HeyGen home → Video Agent. Choose the avatar and voice, brand system (or colors in the prompt), chat mode, ~3-minute target, landscape. Turn Seedance off for a traditional training look.
  3. Attach the PDF and adapt this prompt (quoted from the video; one caption word is unclear and marked):
Create a concise and [unclear in captions: "fully"] training video for new retail
associates at North Star Outfitters using only the attached PDF as the factual
source. Cover the service standard, standard return requirements, damaged product
process, manager escalation triggers, practice scenarios, and final employee
checklist. Do not invent, remove, or change any policy. Use a professional,
friendly presenter and a clean corporate style using navy, teal, white, and small
gold accents. Use the avatar for the introduction, section transitions, and final
recap. Use simple retail b-roll and motion graphics for the policy steps and
scenarios. Keep each scene focused on one idea. Use a confident, conversational
tone and target a run-time of approximately 3 minutes.
  1. Review the scene plan section by section against the PDF; correct it in chat.
  2. Check the credit estimate, then generate.
  3. Run the five checks; edit individual scenes with targeted instructions.
  4. Edit a copy in AI Studio for script, voice, media, text and caption fixes; do a final policy check and preview the narration.
  5. Generate the final version, download, and upload to the LMS, onboarding portal or internal resource library.
  6. Optional: localize the finished video with HeyGen video translation, whose tutorials recommend the localized-accent option for training and onboarding.

Implementation

Tool/Service: HeyGen Video Agent (first draft from a document) plus HeyGen AI Studio (scene-level editing). Setup: a HeyGen account, an avatar, and a structured policy document (PDF). Optional brand system in HeyGen. Cost: credit-based; HeyGen displays an estimated credit cost before generation. No figures stated in the source. Integration notes:

  • Video Agent output opens in AI Studio as a copy, with each generated scene on the timeline.
  • Uploaded company screenshots, videos or document pages can replace stock footage per scene.

Open Questions

  • Video Agent mode names. This source (2026-10-10) says “chat mode” vs “autopilot”. HeyGen Hyperframes (source dated 2026-07-23) names the modes Plan and Autopilot. Possibly a rename or a caption difference; not confirmed.
  • Credit cost for a 3-minute training video is not stated, nor the premium avatar model’s surcharge.
  • The tutorial does not cover SCORM export, quizzes, or LMS-specific packaging; download-and-upload is the only path shown.
  • How strictly Video Agent honors “use only the attached PDF” is shown on one example; the presenter’s mitigation is human review of the plan and the final script.