Green-screen Format reaction clip, revised with previous_run_id

What you get and how to ask for it

sume-green-screen is a Fastlane Format: its catalog description reads "Create a finished Fastlane green-screen Blitz from brand DNA", for a green screen, talking-head overlay or a Fastlane green-screen ad. Call it at POST /v1/formats/sume/sume-green-screen/runs with a key that carries formats:write, and describe the reaction in instruction, for instance a person reacting to a gift guide page behind them.

The description marks it as not for custom generated scenes, social posting or the editor, so keep the brief to the overlay structure and put your brand details in input.

Revise one thing without starting over

A reaction clip rarely lands on the first take. Instead of a new run, send previous_run_id on a new request to the same Format: the next turn continues the same conversation and the agent is replayed what it produced, so it can redo one part and leave the rest alone.

The run you name must be continuable: its receipt shows a thread_id that is not null, and it either completed or left a non-empty artifacts[]. A continuation is a new run with its own id, spend cap and single webhook, and the original never changes.

curl -sS -X POST "https://api.sume.com/v1/formats/sume/sume-green-screen/runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: gift-react-v1-rev-1" \
  -d '{
    "previous_run_id": "arun_REPLACE_WITH_YOUR_RUN_ID",
    "instruction": "Keep everything. Change only the closing line to: Last day is Cyber Monday.",
    "generation_spend_cap_usd": 20
  }'

The refusals that tell you what to do

Four answers cover almost every failed continuation, and each names the fix.

A reaction brief that works

The overlay structure carries the clip, so the brief should be about the reaction. Say who is reacting, to what, and in what mood, in two sentences: for example a shopper at a desk reacts to a gift guide page behind her with a laugh, then points at one item. Name the offer once, at the end. If you need a specific caption style, add the captions later with the standalone Video captions surface rather than asking the run to bake them in.

Because a revision is a new request, it also carries a new generation_spend_cap_usd. Set it to the revision's own budget, not the original run's.

Pricing and bookkeeping

Usage stays per run while artifacts[] on a continued run lists everything the whole conversation generated, so sum usage across the chain yourself. Bind the same output_schema every turn if you use one, because it is not inherited, and use a new Idempotency-Key for each revision.