Input: approved media, script, and bilingual glossary
Output: captioned lesson MP4 in 16:9 or 9:16
Aspect: 16:9 for lessons; 9:16 for social explainers
Build log
1. Lock the picture
Do not pay to rerender media for a wording change. Approve the sequence first.
2. Build a glossary
I keep one approved translation per term. “Score”, “account”, and technical nouns stay identical across lessons.
3. Place captions in the safe area
I test 16:9 and 9:16 crops. Captions never cover the action or the instructor’s mouth.
4. Check with sound off
A learner should still understand the action from captions. If not, the caption has too many words.
Complete execution record
Task boundary and source gate
Treat AI course voiceover and subtitles as a deliverable, not a definition. The job receives approved media, script, and bilingual glossary and owes captioned lesson MP4 in 16:9 or 9:16 in 16:9 for lessons; 9:16 for social explainers. Write the rejection reason first, then turn it into the first hard constraint.
Require the objective verb, terminology sheet, worked example, assessment item, and the source of any claim. If a slide contains two ideas, split the node before spending credits. Do not let a caption introduce a claim that the approved source does not contain. A pretty diagram cannot override wrong terminology.
The imgmov route
Break the outline into Canvas nodes before generating media: one observable objective, one example, one check. Slides, notation, and approved screenshots are references. Subtitles, review questions, and progress markers go to the external editor after picture lock.
WAI describes captions as synchronized text of speech and non-speech audio needed to understand content, and states that automatic captions are not sufficient. Review captions against muted playback before delivery.
Lock the variables that cannot drift
Lock terminology, notation, example numbers, instructor identity, clothing, background light, diagram style, and caption format. A course should feel like one teacher with one method.
Put the locked fields at the top of the prompt, not at the end: Lock media first, then add voiceover, captions, and music in the external editor. Caption work does not consume video-model credits; generation cost comes from the selected video model, such as 10 credits for a 5-second Agnes Video shot. Save that prompt with the reference, ratio, and model so the next run starts from a decision instead of a guess.
Step-by-step build path
1) Lock the picture: Do not pay to rerender media for a wording change. Approve the sequence first. Leave one checkable artifact from this step; do not start the next until it exists. 2) Build a glossary: I keep one approved translation per term. “Score”, “account”, and technical nouns stay identical across lessons. Leave one checkable artifact from this step; do not start the next until it exists. 3) Place captions in the safe area: I test 16:9 and 9:16 crops. Captions never cover the action or the instructor’s mouth. Leave one checkable artifact from this step; do not start the next until it exists. 4) Check with sound off: A learner should still understand the action from captions. If not, the caption has too many words. Leave one checkable artifact from this step; do not start the next until it exists.
Change one named variable per rerun: prompt, reference, camera, duration, model, ratio, or export crop. If two variables change together, a better output cannot be reused because nobody knows which fix worked.
Review gates and evidence
Mute the video and read the captions. Then play it once and ask whether the learner can complete the ending check without rewinding.
Keep source, approved wording, rejected version, correction, credit cost, and final crop next to the asset. The record exists so the next reviewer can reproduce the decision without a chat thread.
Failure diagnosis and retry ladder
If learners fail the check, cut a concept instead of adding a recap. If notation drifts, return to the approved slide. If captions lag, split the sentence at a phrase boundary.
Do not upgrade on hope. A 5-second Agnes Video probe is 10 credits. If the concept passes, Seedance 2.5 costs 90 credits at 480p or 195 at 720p for 5 seconds; Kling is 195 without audio or 245 with audio. Veo enters only when a 4/6/8-second cinematic bucket is genuinely worth 80/125/165 credits. Prove hook, subject, and rhythm first, then pay to clean up motion that already worked.
Versioning and handoff
Deliver classroom, mobile, and captioned versions separately. Save the objective, reference set, prompt, and check as the next module's starting point.
The folder is boring on purpose: source, approved reference, generation settings, caption file, platform cuts, QA screenshots, and one correction line. A reusable asset is boring in the right way.
Why this is not a generic answer
The imgmov advantage is the chain: Asset Library preserves the subject reference, Canvas fixes shot order and first/last frames, the workspace routes Agnes to Seedance, Kling, or Veo only after proof, and your external editor adds captions and CTA without regenerating the media.
Do not let a caption introduce a claim that the approved source does not contain. A pretty diagram cannot override wrong terminology. If the request only asks what ai course voiceover and subtitles means, a search page is faster. This page is useful when someone must deliver captioned lesson mp4 in 16:9 or 9:16 under real constraints.
Frequently asked questions
Does captioning cost video credits?
No. Generation credits are for the selected video model, not editor caption changes.
The workflow is specific about its stopping point: it starts with approved media, script, and bilingual glossary and stops at captioned lesson MP4 in 16:9 or 9:16. In imgmov, upload the source, lock it as a reference, set 16:9 for lessons; 9:16 for social explainers, run one proof, and save the passing settings. If a reviewer cannot tell which version is approved, the workflow has failed even when the media looks good.
Can I add Chinese and English?
Yes. Keep them on separate lines and do not shrink text below readability.
Do not upgrade on hope. A 5-second Agnes Video probe is 10 credits. If the concept passes, Seedance 2.5 costs 90 credits at 480p or 195 at 720p for 5 seconds; Kling is 195 without audio or 245 with audio. Veo enters only when a 4/6/8-second cinematic bucket is genuinely worth 80/125/165 credits. Prove hook, subject, and rhythm first, then pay to clean up motion that already worked.
When should I add voiceover?
After picture lock. Then a wording fix does not require a new render.
The check is not “does it look AI-nice?” It is: Mute the video and read the captions. Then play it once and ask whether the learner can complete the ending check without rewinding. Then the source, approved copy, rejected version, correction, and final crop stay in the same handoff folder.
What should a caption line contain?
One phrase or one clause. Two ideas are hard to read at speed.
If learners fail the check, cut a concept instead of adding a recap. If notation drifts, return to the approved slide. If captions lag, split the sentence at a phrase boundary. Change one named variable, keep the old version, and record the credit cost. If the same failure repeats, fix the reference or scope rather than asking the prompt for forgiveness.
What resolution should I export?
Use the plan-supported output. Guest previews are 1K; paid plans can reach 4K where allowed.
The imgmov advantage is the chain: Asset Library preserves the subject reference, Canvas fixes shot order and first/last frames, the workspace routes Agnes to Seedance, Kling, or Veo only after proof, and your external editor adds captions and CTA without regenerating the media. The handoff is reusable only when the source, reference ID, prompt, model, cost, rejection reason, and approved cut are linked. That chain is what makes the next ai course voiceover and subtitles task faster.