Omni video edit on Sume: the 8-second default holds $1.00 at 720p
Omni edit takes its length from your source clip, but the reserve uses an 8-second hint unless you send duration. At 720p that hold is $1.00 on Sume.

When you send video_url to gemini-omni-flash-1.1 on Sume, the output length comes from your source clip, and the hold placed on your wallet uses an 8-second hint unless you send duration. At 720p that hold is 8 x $0.125 = $1.00.
Sume's docs state this directly: fal does not return an output duration for an edit, so the reserve is a hint, not a measurement of your clip.
What the docs say about edit and duration
In edit mode, Sume sends no aspect_ratio and no duration to the provider, and aspect_ratio returns a 400 if you include it. If you send a duration together with video_url, it is only the reserve-estimate hint, and the default is 8 seconds.
resolution is optional and defaults to 720p. The edit also refuses generate_audio: false, bitrate_mode and reference_audio_urls, because the model always produces native synced audio.
The hold at each resolution
Omni's provider list is $0.10, $0.15 and $0.30 a second at 720p, 1080p and 4K. Sume bills list x 1.25, which is $0.125, $0.1875 and $0.375. Multiply by your hint to get the hold.
The table shows the hold for a few hints at 720p and 1080p. I did not find the way the final capture is computed in the repo docs, so read usage.cost on the finished job, and treat the numbers below as the amount your wallet must have free at submit.
| Duration hint | 720p ($0.125/s) | 1080p ($0.1875/s) |
|---|---|---|
| 3 s | $0.375, rounded up to $0.38 | $0.5625, rounded up to $0.57 |
| 5 s | $0.625, rounded up to $0.63 | $0.9375, rounded up to $0.94 |
| 8 s (default) | $1.00 | $1.50 |
| 10 s | $1.25 | $1.875, rounded up to $1.88 |
Send the hint that matches your clip
If your source clip is 4 seconds, add duration: 4 so the hold reflects the clip, not the 8-second default. It will not change the output length, because the source decides that. It changes only the estimate.
A batch of 50 five-second edits at the default hint holds 50 x $1.00 = $50.00 at submit, while a batch with duration: 5 holds 50 x $0.625 = $31.25 (each job rounded up to $0.63, so $31.50 in practice).
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: omni-edit-hint-001" \
-d '{"model":"gemini-omni-flash-1.1","prompt":"Replace the bottle with an apple. Keep everything else the same.","video_url":"https://example.com/clip.mp4","duration":5,"resolution":"720p","mode":"async"}'Limits to remember
Omni is 3 to 10 seconds, so a longer source is outside the model's range. Check your clip length before you submit. The edit is only on the Video Router surface, not on the OpenRouter-shaped /v1/videos body, and sume/auto does not route to it unless you send video_url.
For prompts that work well for edits, see the car color edit and adding fireworks.
Before you queue a batch
Run one edit, read the finished job's usage.cost, and compare it with the hold. Then size the wallet for the batch from the real number.
- Send
durationthat matches the source to keep the hold honest. - Use a unique
Idempotency-Keyfor each edit. - Keep the source under 10 seconds.
- Do not send
aspect_ratio,generate_audio: falseorreference_audio_urls.
Related posts
More in Pricing
- Omni Flash on Sume: a 4K clip of 3 s is $1.13, 1080p of 10 s $1.88
Gemini Omni Flash 1.1 offers native 4K on Sume's Video Router. A 3-second 4K clip costs $1.13; a 10-second 1080p clip $1.88. How to pick resolution by use.
- Gemini Omni Flash pricing: Google quotes 720p; Sume lists four rows
Google's Gemini API pricing page gives Omni Flash video as $17.50 per 1M tokens, about $0.10 a second at 720p. Sume lists 360p, 720p, 1080p and 4K rows.
- Gemini Omni Flash price per second: 5,792 tokens at $17.50 per million
Google bills Omni video output by the token: 5,792 tokens per second of 720p. At $17.50 per million that is $0.1014 a second, against Sume's $0.125 billed rate.
- gpt-image-1 low, medium, high on GPT Image 2.5: Sume prices
Moving from gpt-image-1 to GPT Image 2.5 keeps low, medium and high and adds xhigh and max. Sume's price per tier at 1024x1024 and what omitting quality bills.
Written by Sume