Multi-style video ad for social media

The person brings a product (its site, or a brief and photos) and where the ad runs. You deliver a 15–30 s 9:16 ad, plus a 16:9, whose look changes once, at the turn: the viewer's world first, then the product's world in a look motion graphics can't do, the same people in both.

This is a treatment of a conceptual product or brand film when a change of look expresses the idea. The steps own the look switch and its staging. Rounds and deliverables follow the core skill. For a music video in many looks with lip sync, workflow-multi-style-music-video.

Steps

1. Facts and brand (guide-pull-product-highlights, guide-capture-a-brand).
2. Script with the turn written in (llm-write-an-ad-script; the hook per llm-write-a-hook; the product's world per llm-pick-a-visual-metaphor): the hook in the first look, the turn on the product's moment (the first bite, the box opened), the payoff and the end card in the second. One look per part, two looks for a 15–20 s ad, three at most for 30 s. Gate: the script, the shot table with a look per shot, and the price.
3. Looks per part (guide-pick-a-stylized-look, Process steps 1–2, and Choosing, Mixing looks): a style frame of the turn's shot in each candidate look. Gate: a pick per part.
4. Cast sheets, one per person per look (nano-banana-character-sheet); real people only from their own photos and with consent (guide-depict-real-people-and-places). Gate: the cast in every look.
5. Stills, 9:16 first (guide-pick-a-stylized-look, Process step 4). Stage the two sides of the turn as the same framing in both looks, so the switch reads as one world changing. Gate: a review_choose per shot.
6. Shots (guide-animate-a-shot, guide-choose-a-video-model): stylized looks on H3 Max 768P, photoreal ones on Kling v3 Pro; anyone who speaks on camera per guide-lipsync-video. One motion test per look before its batch. Gate: each look's test.
7. Voice or music (guide-voice-a-script, A narrator for a short; elevenlabs-generate-music, with the swell on the turn). Gate: in one turn.
8. Base cut (ffmpeg-cut-clips-to-a-voiceover; a music-only ad pins each shot's start), −14 LUFS for the 9:16 (guide-pace-a-short-for-social).
9. Captions and the end card (guide-add-titles-and-motion-graphics for the caption route; hn4a-compositor-brand-ad-titles for the end card and key phrases): the type changes with the look. Gate: the live spec (hn4a-compositor, Live spec, then render).
10. Deliver 9:16 and 16:9 (each its own cut, guide-pace-a-short-for-social).

Choices this workflow makes

• One switch, on the turn. The look changes when the story does (the bite, the unboxing), never on a timer; a free style-flicker transition between the looks is in guide-pick-a-stylized-look (Choosing).
• Same framing either side of the switch, so the change reads as the world transforming rather than a new scene.
• The cast is constant: a sheet per person per look, their defining features in every shot prompt.
• 9:16 is the master: it is a social ad; the 16:9 is its own cut from 16:9 stills.

Dead ends

• None recorded on this workflow yet; the traps it inherits (skin tone under coloured light, extra panels on a sheet, people drifting together) are in nano-banana-character-sheet and style-3d-animation. Record the job's own as it runs.

Worked example

The shape of the demo clip: Acme Co. organic ice cream, two children (invented cast) in a sunlit park. 9:16, 30 fps.

At list prices for a job of this size (two cast sheets per look, about eight stills, about 25 s of Kling v3 Pro and H3 Max in 9:16, a bed): about $6–10, the cut and titles free.