UGC-style ad on Sume: a Format run or the avatar endpoint

For a UGC-style ad, Sume gives you two calls. The avatar endpoint, POST /v1/avatar-1.0/talking-video, renders one presenter saying your script in 4 to 60 seconds. A Format run, such as the catalog's sume-close-camera-ugc, hands a brief to an agent that plans and assembles the cut with the generation tools. Pick the avatar endpoint when you control the words and want a predictable clip; pick a Format when you want the whole ad assembled from a brief.

Sources: Sume's Generate avatar video, Format catalog and Calling a Format.

How the two differ

The avatar route is a model call with a fixed shape. A Format run is one unattended agent turn in a fresh sandbox, ending in a receipt with media and, optionally, JSON in your schema. The Format docs say long-form host video typically finishes in 15 to 30 minutes, so design for async.

Read the Format before you call it

The catalog says to read a Format with GET /v1/formats/sume/{slug} first, which returns its description and the io profile that declares the input and output kinds. Both can be null for older Formats, which means not declared. Do not guess the input keys.

curl -sS "https://api.sume.com/v1/formats/sume/sume-close-camera-ugc" \
  -H "Authorization: Bearer $SUME_API_KEY"

A practical split

Many teams use both. The Format makes a first cut of a new concept; once a hook works, the avatar endpoint repeats it with different scripts and keeps the face constant. Whichever you use, label synthetic presenters honestly and never present a generated person as a real customer.

• New concept, unknown shots: Format run.
• Known script, many variants: avatar endpoint with video_inputs.
• Many recipients: bulk runs with one idempotency key per batch.

Bottom line

Choose on who decides the shots. The more you want to control, the more the avatar endpoint fits; the more you want delegated, the more a Format fits.