Large Presenter Teaching Reels
Teach one useful idea while keeping Madhav visibly present. See the style frames for the look. Current project rules and each clip's acceptance state override this recipe.
Load the hyperframes routing skill and your editorial skill (for example video-use, if installed). The editorial skill owns source/transcript analysis, EDL, pacing, proof, captions and QA. Do not re-ask an approved strategy. Read engine-specific skills before authoring.
Composition
- Genuinely moving, tightly framed head-and-shoulders footage spans nearly the full width of the lower 45-50%. A large box containing a tiny person does not satisfy this.
- Upper region: ONE concrete explanatory visual at a time, such as input/code/output, a changing array, a highlighted resume section, project proof or a before/after.
- Short speech-synchronized captions occupy a stable band between explanation and presenter, clear of eyes, mouth, gestures and diagram labels. Check current platform UI safe zones.
- Use occasional full-frame moving presenter for a question, emphasis or useful close rather than a rigid split throughout.
- Preserve useful original demonstrations. Crop the host separately from the guest where appropriate; never label an enlarged guest Madhav. Prefer a plain crop to unstable hair/hand segmentation.
Suggested 1080x1920 starting geometry: teaching x54/y110/w972/h650; captions y790-930; presenter x36/y970/w1008/h950. Crop without stretching. Check head/chin/gestures and bottom UI. Adapt when diagrams need more space.
Teaching and visual richness
Use question -> plain-English idea -> concrete example -> visible change -> takeaway. Reveal what is being discussed, then hold the result long enough to understand at normal playback speed. Code needs actual output; algorithms need visible state updates; resume advice needs readable redacted sections, not a tiny whole page.
Sparse white screens felt empty in testing: use source-motivated texture, restrained cover-derived accents, purposeful diagrams and a larger presenter. No random scenery, neon, excessive gradients, repeated glass panels or dense text. Richness should improve understanding.
Voice and avatar continuity
- Preserve accepted original speech. New narration uses the presenter's own authorized voice, never someone else's.
- Finalize and verify new audio before avatar generation. Feed that exact approved audio to the chosen existing HeyGen look so voice and lips are not generated independently.
- Honor current engine choice; Avatar V was the preference when supported. Recheck support and costs. HeyGen is for talking avatars, not graphics/music/final rendering. Transparent output is conditional, not guaranteed.
- Treat ElevenLabs 1.2x as a clip-specific request. Measure resulting duration; do not accidentally also accelerate the finished avatar. Never inherit an old slow-down into a fast-paced revision.
- Match perceived loudness, pace and tonal balance at real-to-AI joins; avoid the high-pitched voice jump. Check listening and lip sync explicitly; waveforms/stills cannot certify either.
- Remove arbitrary silent hook tails using word-safe cuts and short audio fades. Finish the thought and complete spoken CTA, then allow a brief deliberate visual/music tail.
- Requested music stays quietly under speech. Approximate 10% is a starting fader setting, not a loudness guarantee. Use licensed audio, speech ducking and restrained SFX at meaningful changes. Do not call a track trending without evidence.
Execution and verification
Record scene engines and final assembly. HyperFrames suits speech-timed arrays, diagrams, code/output and typography; Remotion suits existing React components and footage-heavy assembly. Preserve pins and approved work. Hybrid work exchanges verified rendered intermediates, not HTML imported as React; narration/captions need one owner each.
Inspect one moving presenter-heavy beat and one information-heavy beat before scaling. At phone size check recognizable large face, readable example, lips, purposeful motion, comfortable joins and complete ending. Then verify every changed master: full decode, dimensions/CFR, audio, black frames and captions. Separate technical QA, facts, visual/audio review and user acceptance.
Reuse accepted covers unless replacement is requested. Any outside video you study for layout is not a reusable script, face, voice, username, badge, music license or virality guarantee; never include its footage in deliverables. Saving the skill grants no new spending, likeness-upload, sharing or publishing permissions.