AI video generator for advertising: check a Format's sample first

Before you spend on an ad Format, list the Formats and read two fields on each: io and showcase. io tells you what the Format takes and returns. showcase is a worked example that the Format really produced, verified against Sume's generated-media ledger before it is stored. Neither is a quality guarantee for your product, but both cost nothing to read.

Timing matters. Adobe's forecast, reported by Digital Commerce 360 on 2026-09-29, puts US online holiday sales at $275.1 billion, up 6.7%, so the weeks of ad production are now.

What do io and showcase mean?

The catalog page defines them. Read them from GET /v1/formats, or from GET /v1/formats/sume/{slug} for one Format.

What does null mean?

Both fields are null for Formats saved before registration existed. The docs say that means not declared, not "takes no input". In that case fall back to the Format's description and its page. Do not treat a null showcase as a failed Format.

curl -sS "https://api.sume.com/v1/formats" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  | jq -r '.data[] | select(.showcase != null)
      | [.handle, .slug, .io.output_kind, .showcase.created_at, .showcase.media_url] | @tsv'

How should you use the sample?

Open the showcase for the Formats you are considering, and match the output kind to the placement: a video Format for a clip, an image Format for a still. Then run one item with a low generation_spend_cap_usd and your real packshot. The showcase shows what the recipe can do with its own example input; it says nothing about how your product photo will come out.

For catalog Formats the run, its media and its spend belong to the key that called, and the shared Format stays unowned. That means a test run lands in your history with a receipt, which is the record you compare against the showcase.

A practical order is: list, filter to video or image output, open three showcases, run one cheap test per candidate with the same packshot, then compare receipts. Keep the winner's slug and the instruction text together so the next run repeats the recipe exactly.