Shop Visiting, Seeding, and Sales Short Video Generator

Problem it solves

UGC videos for shop visits, product seeding, and sales often fail at concrete production steps: weak hooks, low information density in 15-second cuts, copy that sounds like ads, creator images with an obvious AI look, and covers or titles that do not match the video. This skill splits the workflow into confirmable stages, reducing rework caused by generating a final video directly from a one-line brief.

How it works

• Scene detection: classifies input as shop visiting, product seeding, or sales based on keywords; ambiguous cases default to seeding with a note that it can be switched.
• Material gating: asks for product, store, or experience assets; if the user proceeds without them, the output is flagged as AI-generated visuals with expected mismatch risk.
• Story and storyboard: produces the core selling point, first-3-second hook, and 0-3s / 3-7s / 7-12s / 12-15s pacing, then a shot table with shot ID, timing, visual description, voiceover, and text overlays.
• Creator image: matches personas such as food explorer, lifestyle deal curator, beauty reviewer, fashion creator, or local spot explorer, then generates 4:5 reference images with real skin texture, phone-shot texture, and lived-in backgrounds.
• Video and packaging: calls the Kling kling-v3 series to generate a default 15-second 9:16 vertical video, plus a cover, three title options, and 5-8 hashtags.
• Confirmation loop: each step waits for user approval; users can regenerate any stage, such as re-creating only the persona without restarting the whole task.

Boundaries and cautions

It fits best for UGC-style short content with real assets or clear product/store context, not for abstract brand films or fully synthetic high-end ads. Without real assets, product, store, and effect visuals may drift from reality. The voiceover is tuned to roughly 75-90 Chinese characters for 15 seconds, so duration changes require rebalancing information density. Creator images, covers, titles, and tags depend on confirmed persona and selling points; skipping confirmation can make later outputs inconsistent.