---

name: "course-studio-render-pack" version: 0.1.0 description: "Rules for producing narrator-driven course videos."

Course Studio Render Pack (opinionated, free)

What

The production rulebook for narrator-driven educational course videos: the format, the visual laws, the text-box discipline, and the QA gates that make a video shippable. Built from dozens of real production cycles — every rule below exists because breaking it cost a re-render. An agent reading this produces videos in the house style on the first pass.

When to use

• Rendering a new lesson for a video course in this format.
• Reviewing/QA-ing a finished lesson before delivery: the gate at the end is the verdict, not a checklist decoration.
• Starting a NEW course: read the "new-course standard" section first — rules ratchet up between courses, never down.

How

1. The format (non-negotiable)

• Narrator-driven. Characters never speak. Scenes show realistic, diverse, real people in complete environments; a narrator describes the scene and the challenges. No character dialogue, no mouth movement, no lip-sync — ever.
• Professional 2D graphics. Polished, never childish (no stars, no hearts, no cartoon props).
• Full-environment realism. Every scene has a FULL environment — objects in place, correct perspective, correct scale. A character never floats in a void or against a flat backdrop.
• Real lighting, weathered people. Light effects are real: neon flickers with irregular stutter, emergency lights wash red/blue across windows and pavement — never a generic glow overlay. People look lived-in: worn hands, tired faces, working clothes. Never stock-photo polish.
• Living details. Scene-appropriate craft that proves real work went in: rain, car lights, monitor flashes and beeps, ambient motion — all under the narrator, never competing with it.
• Audio: one continuous gentle theme per video, ducked under narration, no long silent gaps. Never reuse the same background track across videos — each video gets its own calm, appropriate theme. Scene background music plus ambient effects as a standing rule.
• Unique filenames. Never reuse a filename for a reworked deliverable — a rework gets a new revision number (-rev<N>). Filename reuse caused a real incident where a stale cached copy shipped instead of the fix.

2. Text-box law (the most rejected rule — read twice)

• Teaching points go in text boxes placed in non-intrusive frame areas that block no character activity.
• Boxes cover every lesson point and fill the video — never sparse, no fixed count.
• Each box is synchronized with the narrator: it enters with the words and stays until the point finishes.
• No text may ever exceed box margins at full resolution — wrap inside, and re-verify after any translation (translated text is longer; Spanish reflows must be re-checked).
• Every Knowledge Check question AND its answer is narrated aloud and boxed.
• The QA that matters: overlap. Check that no two text boxes overlap each other, and that no box collides with on-screen citations. Presence checks ("14 of 14 boxes in frame") are worthless without the overlap check — a probe that only counts boxes will pass a broken video.

3. Voice discipline (multilingual editions)

• One consistent, vetted voice per lesson per language. A narrator switching mid-lesson (or a sentence starting clipped) fails the lesson.
• QA the final mix, never the build inputs. A corrected narration track sitting on disk while the shipped mix uses the stale track is a real failure mode. The gate listens to the delivered file.

4. Disclaimers and links

• Disclaimers: full disclaimer in the first video of a course, a short update in a middle video, full in the ending video; every other lesson carries none.
• Validate every link before rendering. All text links in scripts and resource files are checked live as active before the render starts — a dead link in a shipped video is a credibility defect.

5. Render hygiene

• Heavy renders go to the most powerful available machine by default (never the weakest box out of habit).
• Never put a short timeout on a full encode pass; encodes run far slower than realtime under load.
• Template discipline: when a lesson is approved as the style template (e.g. "lesson 1.1 is the template for all future videos"), every later lesson is measured against it for style, narration, and effects.

6. Delivery gate (all must be true — fresh watch, final mix)

• Fresh end-to-end watch of the delivered file (not the build inputs)
• Box overlap checked: no two boxes overlap, no box/citation collisions
• One consistent voice per language, no clipped starts
• Every Knowledge Check question and answer narrated and boxed
• Text inside box margins at full resolution (post-translation reflow re-verified)
• Phone playability proven (plays on a real phone, not just desktop)
• Unique revisioned filename; delivered to the approver in chat — a file staged on disk is not delivered

References

• Template lesson: whatever the current course's approved 1.1 is — style, narration, and effects measured against it.
• Render stack: ffmpeg pipelines with per-lesson build scripts; validate links and disclaimer placement before the encode starts.