Podcast teaser video: Omni cover-art clip plus your audio

To make a podcast teaser, animate the cover art with Gemini Omni Flash 1.1 for the picture and take the voice from your own recording, because Omni cannot use an audio file you upload. Google's Omni page, read 2026-10-08, says uploading audio references is unsupported in the current API, and the Sume catalog agrees: reference_audios is false for gemini-omni-flash-1.1. A 9:16 clip of 8 seconds at 720p costs $1.00 on Sume.

So the job has three parts: a moving image, the audio you want to be heard, and a join. Sume has a step for each.

Step 1: animate the cover art

Send the square or portrait cover as image_url. Omni accepts 16:9 and 9:16 only, so choose 9:16 for a vertical teaser, write the camera move in the prompt, and check the framing of a square cover in a 360p draft.

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: podcast-ep42-teaser-01" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Slow push-in on the cover art, soft studio light, no dialogue",
    "image_url": "https://example.com/ep42-cover.png",
    "duration": 8,
    "resolution": "720p",
    "aspect_ratio": "9:16",
    "mode": "async"
  }'

Step 2: pull the audio you want

If the best 15 seconds already exist in a recorded episode video, audio detach extracts the track of one Sume-hosted video as a wav or mp3, with an optional range. If you only need a cut of the video, video trim returns [start, end) as a new MP4. Both take a media.sume.com URL, so import the file first with POST /v1/media-imports.

The audio file can then be the audio.url of a Timeline render. The audio spine, in Sume's wording, stays on Timeline 1.0. The Omni clip carries its own sound that cannot be switched off, so decide whether the Timeline audio replaces it.

What each piece costs

Only the Omni step has a published per-second price in the pages used here, so the table shows that step against the nearest Google direct list.

Limits to check before you publish

Omni outputs carry Google's SynthID watermark, which Google describes as invisible. A face on the cover art that belongs to a recognizable person may be refused: Google says uploading images with certain recognizable people is unsupported, and filters apply to prompts and output. Run a 360p draft first at 30 percent of the 720p rate, check the motion, then pay for the final.

Where your own audio goes

Omni takes no audio file as input, so the spoken episode audio cannot drive the clip. Make the 10-second visual first, then mix the real audio in your editor. Sume's audio-detach and timeline audio steps are documented for separating and placing audio on a finished video, which is where the episode excerpt belongs.