Iterate AI video prompts one change at a time on Omni 360p

Change one element per run, keep a log, and run the drafts at 360p. OpenAI's Sora guide recommends changing one element at a time, and Google's Omni 1.1 Flash post describes 360p as a draft mode up to 60% faster and one-third the cost of 720p. Together they give a cheap, readable way to tune a prompt on Gemini Omni Flash through Sume.

Sources: OpenAI's Sora 2 Prompting Guide, Google's Omni build post and the Gemini API pricing page (read 2026-10-03).

A log that makes runs comparable

Fix the seed-like parts: subject, duration, aspect ratio and resolution. Change one thing: camera wording, light, or action. Store the job id with the changed phrase. Without that log, two renders differ in too many ways to learn anything.

Budget

Google lists Omni at about $0.10 per second at 720p, so six 3 second runs at 720p would be 18 x 0.10 = $1.80 at Google's list price and $2.25 on Sume at list x 1.25. At Google's description of 360p as one-third of the 720p cost, the same six drafts are about $0.60 at Google's list. The Sume 360p rate is in the live catalog; read it there.

One run

Use a new Idempotency-Key per run, and reuse the same key only to retry the same request.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: iter-run-02" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "In a single continuous shot, push in on a ceramic mug on a desk. Soft window light, warm lamp fill.",
    "resolution": "360p",
    "duration": 3,
    "aspect_ratio": "16:9",
    "mode": "async"
  }'

Promote the winner

Rerun the best draft prompt once at 720p. Google describes 1080p and 4K as upscaled outputs, so pick the final resolution last. See the Video Router docs.