Post-Render QA Checklist for AI Explainers Before You Publish
Catch AI video failures before the feed: lip/face drift, text illegibility, claim mismatch, crop, mute captions—and who signs off.
Post-Render QA Checklist for AI Explainers Before You Publish
Run a practical gate on every AI explainer—lip/face drift, text illegibility, claim mismatch, crop, mute captions—against the approved board before it hits the feed.
Small teams shipping AI image/video explainers lose trust when a pretty render contradicts the storyboard. Memory of the brief is not a QA oracle. Pair this gate with common regeneration-forcing mistakes and accessibility for educational Reels.
QA against the board, not against memory
Open the approved board beside the render. For each beat, ask:
- Does the visual match the marked type (UI / B-roll / card)?
- Do on-screen words match the claim spine?
- Would a mute viewer still get the mechanism?
If you QA from vibes—"looks fine"—you will ship face drift and claim stretch. Stakeholder rituals help; see storyboard review rituals.
The gate (copy into your tracker)
| # | Check | Fail looks like | Action |
|---|---|---|---|
| 1 | Face / identity lock | Host morphs mid-clip; age/gender drift | Regenerate host beats only |
| 2 | Lip / mouth sync (if talking head) | Mouth moves without speech or vice versa | Re-render VO-linked beats |
| 3 | Text illegibility | Captions unreadable at phone size | Enlarge / re-place; check contrast |
| 4 | Claim mismatch | Render implies stronger claim than board | Fix captions or kill clip |
| 5 | Crop / safe zone | Key UI under platform chrome | Re-crop; do not stretch |
| 6 | Mute captions | Mechanism only in VO | Add on-screen rule |
| 7 | UI version / accuracy | Old chrome or invented controls | Replace proof beats |
| 8 | Continuity | Wardrobe/prop flicker across cuts | Local regen or cut |
| 9 | Audio | Music fights VO; silence where cue needed | Remix levels |
| 10 | CTA honesty | End ask not boarded | Align to soft CTA beat |
Do not treat the list as theater. A single fail on claim mismatch or UI accuracy is a hard stop.
Who signs off
| Role | Signs | Does not own |
|---|---|---|
| Producer / content ops | Checklist complete vs board | Legal nuance alone |
| Subject-matter owner (PMM) | Claim accuracy | Type kerning debates |
| Brand / IP owner | Host + wardrobe lock | Platform crop quirks |
| Publisher | Final mute pass on phone | Inventing new claims at upload |
One person can wear multiple hats on a tiny team—but the roles still need explicit ticks. "We all glanced" is how drift ships.
Severity triage
| Severity | Examples | Ship? |
|---|---|---|
| Blocker | Wrong claim, fake UI, unreadable mechanism | No |
| Major | Face drift on hero beat, CTA mismatch | No until fixed |
| Minor | 2-frame flicker, music bed slightly hot | Fix if cheap; else note |
| Nit | Preferred synonym in caption | Do not delay publish |
Regenerate locally. Full-clip rerolls for nits create prompt debt all over again.
Phone-size mute pass (non-negotiable)
- Export or preview on a phone
- Watch muted once
- Watch with sound once
- Screenshot frame 1 and the mechanism beat; confirm readability outdoors-ish brightness
If muted pass fails, do not "fix it in the caption field of the social post." The video itself must teach.
Sample QA log row
| Field | Example |
|---|---|
| Asset | invites-viewer-role-v3.mp4 |
| Board | invites-viewer-role board v2 |
| Reviewer | Producer + PMM |
| Blockers | none |
| Majors | caption contrast on beat 3 — fixed v3.1 |
| Mute pass | pass (phone) |
| Publish | yes / 2026-10-30 |
Keep the log beside the board. Future you will need it when UI changes mid-campaign and someone asks why an old chrome still shipped.
Automation vs human judgment
You can automate file-naming checks and ratio checks. You cannot automate "does this claim overreach?" or "is the metaphor still mapped?" Budget human minutes for blockers 4 and 6 on the gate. Everything else can be assisted—but signed by a person.
Edge cases worth a dedicated glance
| Edge case | What to check |
|---|---|
| Multi-ratio exports | Same claim spine; crops not stretched |
| Translated captions | Claim meaning preserved; no soft-Scro product rename errors |
| Loopable endings | Last frame does not contradict first-frame promise |
| Soft CTA | Ask matches boarded job; no surprise hard sell |
| Metaphor → UI cuts | Mapping card present before chrome |
Time budget
For a 45s teach, a disciplined QA pass is often 8–12 minutes once the board is open beside the player. Skipping it to "save time" is how you spend an afternoon on comments and regenerations after publish. Budget QA as a first-class beat of the production schedule—same as render.
Pre-flight vs post-render (do not conflate)
Pre-flight (before render): board approved, stills locked, claims signed, style lock present.
Post-render (this article): the checklist above against the actual pixels and audio.
Teams sometimes "QA" by re-reading the brief and skipping the player. That is pre-flight nostalgia. Post-render means play the file.
Red-team questions (2 minutes)
Ask a teammate who did not board the piece:
- What is the transferable rule?
- What would you click in-product tomorrow?
- Did anything look fake or overclaimed?
If they cannot answer (1) on mute, fail the mute captions check. If (2) points to a path not shown, fail claim/UI alignment. If (3) flags face drift or invented chrome, fail identity/UI checks.
Soft close
QA against the approved board, not against memory of the brief. Storyboard-first workflows (Scro) make the board the contract your checklist grades against.