Creator-explainer style (opt-in)

Use this style ONLY on the video where the user asks for it ("is style mein edit karo", "behind-me captions wala style", "reference shorts jaisa", "creator-explainer style mein"). For every other edit, use short-form-video-editing.

Everything not covered here follows short-form-video-editing: delivering the complete video, the face-hiding flower, transcription, music synthesis, export size, shell pitfalls, the Golden Rule and the language rule (Roman Urdu, "ap").

This style was learned on 2026-10-15 from 4 creator shorts the user supplied (creator-income and AI-editing explainers).

• Reproduce the same placements, B-roll types, half-screens and sound effects, but with original visuals for each video. Never copy the reference frames.
• The fonts are provisional until the user names their own lettering.

1. Talking head + text placements

The speaker is centred in a dark, moody room with coloured practical lights. The edit cuts away often and returns for punchlines and numbers.

A. Subtitle pills (every spoken phrase, for the whole video):

• A white rounded rectangle (#F4F4F6, radius ~12px, padding ~12 × 24px) with dark text #111.
• A semibold geometric sans (Poppins / Plus Jakarta Sans 600), ~38–40px, 2–5 words per pill.
• Centred at y ≈ 1530 on every shot.
• Animation: only a 0.1 s fade/scale 0.96 → 1.

B. Key captions BEHIND the person (the user's favourite; they must look clean and professional, never cheap):

• Serif + bold: a small serif line (DM Serif Display / Lora, white, 64–84px) over one big bold word (Inter 800, 150–250px, white → lavender #C8C8F8 gradient, letter-spacing −0.03em).
  • Place it across the chest (y ≈ 1100–1250) with the hands/mic in front.
  • Or place it huge behind the head/shoulders (y ≈ 150–450) for section words ("1st", "2nd").
• Word-highlight: a bold rounded sans in white, with the current word in mint #08D080, word by word at chest level (y ≈ 1050–1150).
• Motion: words blur in (8px → 0) and rise 20px over ~0.3 s, staggered, then fade out fast.

C. Glowing stat text (every number spoken):

• "That's about" (Inter 600, ~70px) above a big number (Inter 800, 130–170px).
• White, with a soft white + light-blue glow.
• On the chest at y ≈ 1150–1280, usually behind the hands.
• The number counts up for 0.6–1.0 s to the real spoken value.

D. Mint keyword pill: #48E880 with 1–3 white bold words, on a split seam or the chest. It pops in over 0.15 s (scale 0.8 → 1).

E. Floating cards:

• Screenshots/product photos with a 2px brand border (mint #48E880), radius ~16px.
• Place them over the lower body (y ≈ 1170–1640), or BEHIND the head at the top (y ≈ 100–600, with the person in front).
• Use side-by-side pairs with labels for comparisons.

F. Emphasis moves:

• Viewfinder corner brackets around the face.
• A quick exposure flash (+30% for 0.2 s) on a punch word.
• 4–6% punch-ins on hook words.

2. B-roll and full-screen graphics

These fill ≈ 55–65% of the runtime, with a new visual every ~2–3 s.

Transitions: mostly hard cuts on the beat; a pixel-mosaic dissolve on the hook; circle wipes; 3D camera moves; blur-pushes.

3. Half-screen layouts

• Split hook:
  • The top ~60% is a designed title card: a navy #081860 → #040A2A gradient, large glowing arc lines, two glass icon chips, a two-tone title (white + sky blue #58B8E8), a dark pill, and a small stat card with a blue icon dot.
  • The speaker sits below, in a card with rounded TOP corners (~70px) and a glowing sky-blue top edge.
• 50/50 hard split:
  • Two full-width halves (0–956 / 964–1920), a 6–8px white seam, and a mint pill centred on the seam.
  • Alternate which half holds the speaker.
• Both halves UI: two screens, with the speaker as a small PiP inside one.

4. Sound (measured from the references)

• Music: an original beat-driven bed throughout.
  • ≈ 110–120 BPM and kick-led for tech/explainer videos; a culturally fitting groove for product/fashion.
  • About 16–18 dB under the voice, ducked under speech, lifting in the gaps.
• Sub-boom / low impact on every big reveal, every number reveal and every new B-roll type (~6–10 per minute).
• Whoosh / swish on camera moves, 3D pushes, text slides and card entrances.
• UI click + bright ding on cursor taps and buttons.
• Digital crackle under the pixel-mosaic transition, and optional soft ticks under count-ups.
• Mix: stereo, panned with the motion; master ≈ −14 LUFS, peak ≤ −1 dBFS; the voice always on top.

5. Captions behind the person (procedure)

1. Install rembg with pip install --break-system-packages "rembg[cpu]" and use the session new_session('u2net_human_seg') (the model comes from GitHub releases).
   • It takes ~1 s per 1080x1920 frame on CPU, so run it only for the frames that carry behind-captions or behind-cards.
2. Use only_mask=True, feather 1–2px, and build RGBA person frames for those ranges.
3. Layer order: graded video → caption/card → person cut-out → face flower (if hiding a face) → subtitle pills.
4. If the person fills the frame width, put the text behind the head/shoulders at the top rather than the chest, so part of it stays readable.

6. Analyzing new references for this style

• 1-fps contact sheets with timestamps.
• Full-res key frames at layout changes.
• Colour sampling of the saturated pixels.
• A cut list from HSV histograms.
• Spectrograms with cut markers, and band-energy peaks (30–110 Hz booms, 5–12 kHz whooshes) to map the SFX.