YouTube Create Omni lighting options: reuse them as prompt words

YouTube Create gives Gemini Omni four lighting options: Soft morning light, Golden hour, Blue hour and Moonlight. They are chips you tap, not API fields, so to get the same look elsewhere you write the phrase into the prompt. On Sume that means a prompt string on a 9:16 gemini-omni-flash-1.1 request.

The facts come from YouTube's help page Create with AI in the YouTube Create app, read on 2026-10-02, plus the Video Router docs. YouTube calls these features experimental and available to a limited number of creators.

What does YouTube Create's Omni feature give me?

The help page says Gemini Omni in YouTube Create generates vertical clips of up to 10 seconds, with the four lighting options above. Aspect ratio is 9:16 by default. It is on mobile only, in Australia, Canada, India, New Zealand and the United States, and the prompt limit is 900 characters. Prompts must follow YouTube's Community Guidelines, and the app rejects prompts that break policy or touch sensitive topics. The page also says labels tell viewers that the content is synthetic.

Other YouTube Create AI tools have different footprints: the page lists Edit with AI in a longer list of countries. Do not assume that a country with one feature has the other.

How do I write the lighting into a prompt?

A chip becomes a clause. Keep the scene first and the light second, and name the light the way YouTube names it, so you stay close to what the app would send. Four examples follow.

• Do not stack two lights in one shot. Pick one clause per clip.
• Ask for one continuous shot if you do not want cuts, as the Gemini API guide advises.
• Put negatives in the prompt text, for example no dialogue, because the API has no negative prompt field.

Does a 900-character prompt cap matter on Sume?

The 900 limit is YouTube Create's, and the Sume docs I read do not state a prompt length for this model, so I will not give one. The practical advice is the same on both: say the subject, the action, the light and the sound in a few sentences and stop. Long prompts do not make a 10-second clip richer, because there is only so much time for events to happen.

How do I get the same 9:16 clip from Sume?

Set aspect_ratio to 9:16 explicitly, pick a duration from 3 to 10 and send the request as an async job. Native audio is always on for this model. Use a 360p draft to test the light, then re-run at the resolution you need, since resolution runs from 360p to 4K.

You will not get YouTube's label by calling an API, and Sume does not publish to YouTube from this route. Disclosure and upload are your steps.

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: golden-hour-short-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "A skateboarder rolls past a bakery in one continuous shot. Lit by golden hour sun, long warm shadows. Calm background music.",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

What is the difference in practice?

YouTube Create is a phone editor with a timeline around the clip, a label and a publish path. A Sume call is one clip and a job id. If you want the clip inside a longer Short, generate it, then place it with Timeline, which assembles ordered video slots with an audio spine into one MP4.

How do I test a lighting clause cheaply?

Run the same prompt four times with one clause changed, at 360p and 5 seconds, and compare the clips side by side. Leave the rest of the prompt untouched. Expect each run to differ a little even with identical text, so judge the light, not the exact framing.

Google's launch post says 360p drafts render up to 60 percent faster at a third of the cost of standard 720p, so four test clips cost roughly what one or two full-size ones would. When one clause wins, re-run that prompt at 720p or 1080p. Submit each test with its own idempotency key, so a retry cannot bill the same clip twice.

If you need a look that none of the four options names, such as neon or overcast, write it in plain words instead. The four chips are YouTube's menu, not a list of what the model understands, and the Gemini API guide shows free-form audio and timing instructions working the same way.