A complete faceless video is not merely an exported MP4. It is a chain of decisions that can survive review: a focused brief, a sourced script, authorized voice and visuals, a clean edit, accurate captions, the correct publish settings, and a useful retrospective.

In this capstone, you will run that chain from start to finish. You will publish one video at an appropriate visibility level and assemble an evidence pack containing the brief, sources, prompts, rights records, script, export details, publish record, and retrospective. The pack proves how the video was made; it does not claim the video will perform.

// concept

Lock the Brief, Evidence, and Rights Before Production

Start a folder named video-01-topic-date. Add subfolders for 01_brief, 02_sources, 03_script, 04_voice, 05_visuals, 06_edit, 07_export, and 08_publish-retro. Use filenames such as script_v01.md and script_v02_approved.md so a reviewer can identify the version actually used.

Write this one-page production brief before opening a generator:

Brief field	Decision to record
Audience	One specific viewer and what they already know
Viewer problem	One question this video will answer
Deliverable	Platform, horizontal/vertical format, target length, language
Evidence boundary	Sources the script may use; claims it must not make
Voice	Your recording, authorized clone, licensed synthetic voice, or approved narrator
Visual system	Self-shot, licensed stock, generated scenes, screen recordings, or graphics
Disclosure	What the platform and viewer need to know about realistic synthetic media
Approval	Who signs off the script, voice, visual rights, and final export

Every factual sentence in the script needs a source ID. Keep a source log even for a 45-second video:

ID	Claim or asset	URL/file	Owner or licence	Consent/receipt file	Approved use
S01	Factual source	Official URL	Publisher	Not applicable	Paraphrase
V01	Product close-up	`visuals/phone-shot-01.mp4`	Creator	`creator-release.pdf`	This video, organic social
A01	Narration voice	`voice/voice-sample.wav`	Narrator	`voice-consent.pdf`	This video only
M01	Music bed	Licensed library URL	Composer/library	`licence-receipt.pdf`	Platform publish

“Found online” is not a licence. Record the exact licence or written permission and its date. Do not clone another person’s voice, animate their face, or upload private client footage without explicit authorization.

// concept

Build the Script, Voice, and Shot Plan as One System

Use the brief and source log to draft, then read the script aloud with a timer. A useful faceless script marks what the viewer hears and sees at the same moment.

// prompt — copy me17 lines

Write a 55-second faceless educational video script.

Audience: Pakistani freelancers sending a first video draft to a client.
Goal: teach a three-file delivery package: preview, final export, and rights log.
Tone: calm editor coaching a beginner; plain English.
Evidence rule: use only facts inside <sources>. If a claim is unsupported, write [SOURCE NEEDED].
Structure:
- 0–3s: precise problem-led hook, no exaggerated promise
- 3–38s: three steps, one sentence each
- 38–50s: a quick verification check
- 50–55s: invitation to save the checklist, not an engagement demand
Output a table with: timecode | narration | on-screen text | visual | source ID.
Keep on-screen text under eight words per card.

<sources>
[Paste redacted source notes and IDs here]
</sources>

After drafting, circle every number, comparison, quotation, and current tool claim. Match each to a source ID or remove it. Then expand abbreviations, spell difficult names phonetically, and break long sentences at natural breaths.

For voice, choose one ethical route:

Record your own narration on a phone in a quiet, soft-furnished room.
Use a synthetic stock voice under its current licence.
Use an ElevenLabs voice you are authorized to use. For cloning, save the speaker’s consent and follow the current verification and terms; access can depend on plan and region, so check the official page.

Generate a short voice test first. Check names, Urdu words, currency, pace, and emphasis. Test Roman Urdu as complete sentences. Regenerate only faulty lines, then save the approved audio.

Turn the approved script into a shot manifest before generating visuals:

Shot	Time	Purpose	Source/generation prompt	Rights ID	Status
01	0:00–0:03	Show the delivery mistake	Screen mock-up made for this lesson	V01	Approved
02	0:03–0:12	Explain preview file	Close-up, neutral desk, export window	V02	Generate/test
03	0:12–0:25	Explain final export	Timeline and file labels	V03	Approved

If using Veo, write one prompt per shot rather than one prompt for the whole video. State framing, subject, action, camera movement, setting, light, visual style, and exclusions:

// prompt — copy me4 lines

Vertical 9:16 medium close-up of hands organising three clearly separated video-delivery
folders on a plain desk; slow overhead push-in; soft daylight from camera left; realistic,
clean documentary style; six-second shot; no faces, no brand logos, no readable personal
information, no extra fingers, no floating interface text.

Keep a backup for every essential shot. Because plan gates and controls change, confirm them in the current official guide. Free alternatives include self-shot B-roll, simple animated text, or authorized screen recordings.

// concept

Assemble, Caption, and Pass the Master Gate

Create a CapCut project in the required aspect ratio. Import the approved narration first, place it on the main audio track, and lock it. Add visuals against the voice, then remove any shot that does not clarify the spoken point. Use music only if its licence covers the intended publication; keep it low enough that every consonant remains understandable.

Generate captions from the narration. In CapCut desktop/web this is generally under Captions → Auto Captions → Recognise/Generate; confirm the current location because availability varies. Correct names, Roman Urdu, punctuation, timing, and line breaks manually. Watch the video muted: captions and visuals should still communicate the sequence.

Run these gates on the final timeline:

Gate	Pass condition	Evidence to save
Script	Claims match S-IDs; no unsupported promise	Approved script and source log
Voice	Authorized, intelligible, correctly pronounced	Consent/licence and audio file
Visuals	Every clip has a V-ID; no accidental logos/private data	Shot manifest and licences
Edit	No dead frames, broken cuts, flash frames, or clipped audio	Full-screen review note
Captions	Words, timing, names, and Roman Urdu manually checked	SRT/TXT or caption screenshot
Disclosure	Realistic synthetic material assessed under platform policy	Yes/no decision and reason
Export	Correct aspect ratio, no watermark, complete playback	Export settings and file hash/name

Export a low-resolution review file before the master. For YouTube, the current official guidance supports MP4 with H.264 video, progressive scan, and the source frame rate; check its current resolution-specific recommendations. Watch the export completely on phone and laptop. Timeline playback is not an export check.

// concept

Publish Deliberately and Write the Retrospective

Prepare a platform-native publish sheet: final title, description, thumbnail/cover, caption, audience setting, visibility, source credits required by licences, and altered/synthetic-content decision. Upload privately or unlisted first when the platform permits, process the file fully, and check picture, sound, captions, crop, and cover on a phone. Then move to the intended visibility only after the gate passes.

On YouTube, the upload Details include an altered-content decision for realistic or meaningful synthetic content. Read the current policy; “faceless” does not remove disclosure duties. Choose the audience setting truthfully and avoid unsupported title or thumbnail promises.

Save the publish URL, publish date/time, visibility, final title, cover, and disclosure decision in 08_publish-retro/publish-record.md. Analytics are observations, not proof of why viewers behaved a certain way. Write the retro after the production memory is fresh:

// template — copy me22 lines

# Video 01 Retrospective

**What worked in the workflow**
- [One specific step that prevented rework, with file or timestamp]

**What broke or slowed production**
- [One specific failure, its cause, and how it was detected]

**Rights or disclosure decisions**
- [Asset/voice ID, decision, and evidence file]

**Final quality notes**
- Audio:
- Visual continuity:
- Captions:
- Mobile playback:

**One change for video two**
- I will change [one production variable] because [evidence from this run].

**Do not change yet**
- [An untested assumption that needs more videos or evidence]

// worked_example

Worked Example

This hypothetical sample is not a real campaign or performance report. A Lahore freelance editor plans a 45-second vertical video, “Three files in a clean client handoff,” for beginner phone-based editors. Only the editor’s own checklist is allowed as evidence.

The editor uses the script prompt from this lesson. Part of the sample output is:

Timecode	Narration	On-screen text	Visual	Source
0:00–0:04	“A final MP4 alone can leave your client guessing.”	Final file is not the handoff	Sample folder with one MP4	S01
0:04–0:16	“First, send a small watermarked preview for comments.”	1. Review preview	Phone opens `project_preview_v02.mp4`	S01
0:16–0:29	“After written approval, send the agreed final size and format.”	2. Approved master	Export settings, private details blurred	S01
0:29–0:41	“Include a rights log for music, stock, voices, and generated shots.”	3. Rights log	Four-row sample rights sheet	S01
0:41–0:45	“Save this three-file check for your next delivery.”	Preview · Master · Rights	Three checked folders	S01

Draft one fails the master gate. At 0:17, a screen recording reveals a client phone number; the auto-caption writes “PKR” as “peaker”; the music receipt permits personal use only; and a realistic generated office shot has no disclosure decision recorded. The editor does not blur and publish in a rush.

The editor replaces the screen recording with a fabricated sample interface, corrects the caption, substitutes authorized music, and logs the generated shot. After checking the platform’s disclosure policy, the editor records the decision and reason. The reviewed MP4 is uploaded unlisted, checked on a phone, and moved to the intended visibility. The retro sets one change: build the rights log while sourcing, not after editing.

// failure_cases

Failure Cases to Diagnose

7 cases to diagnose

The script outruns its sources
numbers or product claims have no S-ID. Remove them or attach an official source before recording.
The voice is usable but unauthorized
a convincing clone exists without a signed permission scope. Delete it, document the issue, and record your own voice or use a properly licensed voice.
A generated shot breaks continuity
the same object changes colour or position between cuts. Regenerate that shot with a tighter continuity description, use a still crop, or replace it with simple B-roll.
Captions look finished but are wrong
Urdu names, PKR, or tool names are misheard, or timing covers a visual label. Correct the text and timing manually and perform a muted review.
The editor passes but the export fails
black frames, missing fonts, clipped audio, or a watermark appears only in the MP4. Review the exported file, not just the timeline, and re-export after fixing the responsible layer or asset.
Publication loses the intended frame
a vertical cover or caption is cropped by the platform interface. Preview on the destination app and move essential text inside safe margins.
The retro makes a causal claim from one upload
“the hook caused low reach” is not established by one result. Record the observation, propose one change, and test it without promising improvement.

// pakistan_angle

Pakistan Angle

USD subscriptions can stack into a serious PKR production cost, and Pakistani cards do not always work with every international vendor. Choose the one paid tool that removes your actual bottleneck, check its current official billing page, and keep a no-subscription route: your own phone narration, self-shot B-roll, rights-cleared assets, and CapCut’s currently available free editing features. Agree client payment milestones, transfer fees, revision limits, and who bears paid-asset costs in writing. JazzCash or easypaisa may suit a local client, while an overseas client may require another mutually available method; verify current availability instead of promising one rail.

Load-shedding and unstable mobile data should shape the production schedule. Keep source media and the CapCut project on local storage, create proxy or low-resolution review files, pre-render before an expected outage, and queue the final upload when power and bandwidth are stable. Do not repeatedly upload a large master to collect comments: send a compressed watermarked preview first, obtain written approval, then upload the approved master once.

Roman-Urdu narration needs a human pronunciation pass for names, city references, and mixed expressions such as “PKR five thousand.” Keep alternate phonetic spellings in the voice script. Remove CNIC images, WhatsApp numbers, home addresses, invoices, and client account details from prompts and screen recordings; use invented sample data in demonstrations. Written consent should state whether a person’s voice or footage may be used only in this video or in later client and promotional work.

// hands_on

Hands-On Exercise

8 steps

Produce and publish one 30–90 second faceless video that you are authorized to make.

Create the eight-folder project structure and complete the brief, including audience, platform, format, evidence boundary, voice, visuals, and disclosure decision owner.
Build the source and rights log. Every factual claim, voice, clip, image, music track, and generated shot needs an ID and a permission status.
Draft the timed script with source IDs. Save the prompt used and the unedited output, then approve a corrected script after a timed read.
Produce an authorized voice track and a shot manifest. Save all generation prompts, consent records, licences, and receipts beside the corresponding asset.
Assemble the video in CapCut, manually correct captions, and complete every master gate in the table above.
Export a review file, collect approval if another person owns the brief, then export and watch the final master on phone and laptop.
Upload at a safe visibility, verify processing, sound, crop, captions, cover, audience setting, credits, and synthetic-media disclosure. Publish at the intended visibility only after these checks.
Save the publish record and complete the retrospective with one workflow change for video two. “Done” means the video is published at the chosen honest visibility and the evidence pack contains the approved brief, source log, script versions, prompts, rights/consent records, shot manifest, caption evidence, export settings, publication record, and retrospective. A view count is not part of completion.

// completion_rubric

Completion Rubric

6 checks — tick as you verify

0/6

// sources

Sources

5 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “The script outruns its sources.” The lesson describes it like this: “Numbers or product claims have no S-ID.” What does the lesson tell you to do about it?