Mercado Libre Clips with AI: leave off prices and static images

An AI clip for Mercado Libre Clips has to leave out several things the platform lists as rejected: prices, deals and coupons, static images, third-party music, contact details and watermarks with personal details. Sume's captions step burns text onto a finished clip, so the safest setup is to author the wording yourself and check it against that list.

The rules below are from Mercado Libre's own requirements page, read 2026-10-01. Sume behavior is from the video captions and Timeline 1.0 docs. This is not legal or policy advice; the platform decides what it approves.

What does Mercado Libre say a Clips video must avoid?

The page lists what to skip. These are the items that matter for an AI-made clip:

How do I put approved wording on the clip without a price?

Standalone video captions take a public HTTPS video_url and return a job-backed captioned video. When you omit authored text, captions come from speech, so a spoken price becomes a burned-in price. Pass cues (or segments) with text, start and end instead and the docs say the authored overlay is burned without speech-to-text.

That puts the wording under your control: write each cue, review the list for any number with a currency sign, and only then submit. POST /v1/video-captions is the route, and you read the result at GET /v1/video-captions/:id.

What if the clip has no spoken audio?

Speech-to-captions needs audible speech. A silent clip fails as caption_no_speech with next_action: use_overlay_captions. Mercado Libre wants voice or background music, so a silent product clip needs sound anyway; but if you use authored cues, the caption step does not need speech at all. The related post what to do with caption_no_speech covers the failure.

How do I keep static images out of the edit?

In Timeline 1.0, video[].source_url takes a Sume-hosted clip or a still, and the docs say stills are static holds. A static hold is what the platform's rule describes, so build the timeline from clips only. If you join more than a few segments, render.strategy defaults to auto, which chunks past 12 segments.

The page does not say whether a generated music bed counts as third-party. It does ask for voice or background music, and rejects music belonging to third parties; confirm the rights question with Mercado Libre before you upload.