Module 01 · The AI Video Stack in 2026
Veo, HeyGen, ElevenLabs, and CapCut: Map Capabilities, Not Brands
Open lesson + course map
On this lesson
Course outline
Module 1 · The AI Video Stack in 2026
Module 2 · Scripting for AI Video
Module 3 · AI Voice and Narration
Module 4 · Generative Video With Veo
Module 5 · AI Avatars and Presenter-Style Video
Module 6 · Editing and Assembly in CapCut
Module 7 · Video Service Packaging
An AI-video tool is useful only when it solves a production job. Veo generates shots, HeyGen can present a script through an avatar, ElevenLabs produces narration, and CapCut assembles media—but access and plan gates change.
By the end, you will have a tool-selection matrix and pipeline diagram for one project, with free alternatives, payment triggers and rights checks.
// concept
Map Each Tool to a Production Job
Start with the asset required at each stage. Script writing comes first; these tools enter after the message is clear.
| Tool | Pipeline stage and genuine job | Free-tier reality | Paid trigger |
|---|---|---|---|
| Veo | Visuals: generate shots when filmed or licensed stock cannot express the idea | Verify current: access, country, credits, model, duration and watermark | Generated footage is the tested bottleneck and paid terms permit the use |
| HeyGen | Presenter: deliver exact dialogue through a stock or consented custom avatar | Verify current: region, quota, duration, resolution, avatar and watermark | A presenter is essential, beats B-roll in a test, and permissions exist |
| ElevenLabs | Voice: narrate an approved script with an authorised voice | Verify current: credits, voices, attribution and commercial licence | Voice quality or commercial rights are the bottleneck; free fallback: self-record |
| CapCut | Assembly: cut a timeline, edit captions, mix sound and export | Verify current: platform, region, cloud feature, watermark and export gates | A needed export or collaboration feature is gated; free fallback: its basic editor or another local editor |
Brands overlap. Keep the job stable and re-check official access before each client project.
// concept
Choose a Budget Variant Before Generating
Zero-budget pipeline: write the script → record your voice on a phone → use footage you shot or licensed stock → assemble and correct captions in an available CapCut free workflow. Not every named tool has a usable free tier.
One-paid-tool pipeline: approved script → paid ElevenLabs narration → owned or licensed footage → CapCut. Choose it only if an audition proves narration is weakest and current terms grant needed rights. If visuals are weak, keep your voice and test paid Veo. Pay for the stage that bottlenecks you first.
Use a 10-second proof before buying. In Google Flow, try project → generation settings → Video → available model; confirm current labels and region. In ElevenLabs Text to Speech, select an authorised voice and test one paragraph. Test a HeyGen stock avatar before a custom one. In CapCut Desktop, import to the timeline, then use Captions → Auto Captions → language → Generate; confirm its current location.
// concept
Verify Rights, Consent and Output Conditions
Access does not equal permission. Record tool, plan, date, source, licence, consent, watermark and intended use. Never clone someone because their audio is public. For a HeyGen video-based Digital Twin, the depicted person must complete its consent flow; keep written client approval too.
Review generated faces, hands, logos, packaging and text. Apply required synthetic-media disclosures. Never present a generated market, flood, protest or clinic scene as documentary footage.
// worked_example
Worked Example
The filled selection matrix is:
| Stage | Job | Candidates tested | Decision and evidence |
|---|---|---|---|
| Script | 45-word message | Human draft | Owner approves prices and spelling |
| Voice | Roman-Urdu narration | Owner / ElevenLabs | Owner selected for natural pronunciation and no subscription |
| Visuals | Opening plus owned photos | Veo / stock | Veo only for a non-documentary opener; prompt/date saved |
| Avatar | Presenter | HeyGen / none | None; it would hide the product and add no capability |
| Assembly | 9:16 edit and captions | CapCut Desktop | Selected; captions checked against approved script |
Pipeline diagram:
approved script → owner's recorded voice → Veo opening + owned photos → CapCut timeline → reviewed captions → 1080 × 1920 exportThe opening-shot prompt is specific about composition and excludes unsupported claims:
Vertical 9:16 close-up product-video shot of an unbranded chocolate cupcake
on a plain wooden counter, slow camera push-in, soft warm window light,
shallow depth of field, realistic food texture, no people, no logo, no text,
no packaging, no price labels, quiet neutral background, commercial product style.Sample first-draft observation, not a real result: the clip adds unreadable label text and changes icing colour. Fix the prompt with no label, no text, plain plate, dark chocolate icing throughout; if continuity still fails, use the owner's photo with a CapCut keyframe zoom. In CapCut, place narration first, cut at sentence boundaries, generate captions, correct hazelnut and Roman-Urdu spellings, then export a review copy.
// failure_cases
Failure Cases to Diagnose
6 cases to diagnose
Buying all four subscriptions
produce a short proof first; pay only where the comparison fails.
Using Veo for factual product evidence
packaging, colour or text may drift. Use approved photography.
Adding a HeyGen avatar by default
remove it when B-roll and captions communicate faster.
Cloning a public voice
public access is not consent. Obtain permission or use an authorised voice.
Using free-plan ElevenLabs audio commercially without checking
verify current licence and attribution; self-record if it disallows the use.
Trusting CapCut captions
compare mixed English/Roman Urdu captions with the approved script and listen on a phone.
// pakistan_angle
Pakistan Angle
USD subscriptions change after exchange conversion, card charges and taxes. Confirm current Pakistan support and payment methods; old screenshots are unreliable. JazzCash or easypaisa may not complete a vendor checkout. Record the actual PKR debit, and never send card screenshots or OTPs to a client or reseller.
During load-shedding or limited mobile data, generate short tests, download one take, edit locally or with proxies, and upload on a stable connection. Test Lahore, easypaisa and menu terms before full Roman-Urdu narration. Never upload a CNIC, private WhatsApp number or employee likeness without necessity and explicit permission.
// hands_on
Hands-On Exercise
6 steps
Define one 15–30 second video: audience, format, message and required assets.
Copy the capability table and mark each named tool use, test, or skip.
Fill one row per pipeline stage with the chosen tool, free-tier check date, paid trigger, rights evidence and fallback.
Draw your pipeline as
script → voice → visuals/avatar → assembly → review/export.Run one 10-second test at the stage you are least certain about. Save the prompt/settings and note one defect plus its specific fix.
You are done when another person can follow the matrix without buying an untested tool or using an unapproved face, voice or asset.
// completion_rubric
Completion Rubric
5 checks — tick as you verify
// sources
Sources
5 official sources — check every claim yourself
- support.google.comGoogle Flow Help — supported countries, languages and feature restrictions
- help.heygen.comHeyGen Help — current credit-based plans and free-plan conditions
- help.heygen.comHeyGen Help — recording consent for a video-based Digital Twin
- elevenlabs.ioElevenLabs Documentation — Text to Speech capabilities and usage conditions
- capcut.comCapCut Help — Auto Captions workflow and troubleshooting
// check_yourself
Check yourself
4 questions · answers and options are taken word-for-word from this course
1 / 4 · diagnose
Your work shows this failure mode: “Using Veo for factual product evidence.” The lesson describes it like this: “Packaging, colour or text may drift.” What does the lesson tell you to do about it?