Google Vids Omni edits: color grade and restyle prompts for Sume
Google Vids fixes color grading and restyles video with Omni, but not in the EEA, UK, Switzerland, Texas or Illinois at launch. Try edit prompts on Sume.

Google Vids can edit any video with Gemini Omni using text instructions, such as fixing color grading or restyling to anime, but Google's update says editing non-AI videos with Omni is not available in the European Economic Area, Switzerland, the United Kingdom, Texas or Illinois at launch. The Sume docs for Omni edit list no such region rule, so check your own legal position before you pick either route.
What Vids lists
The Workspace Updates post names the edit examples "fix the color-grading, restyle the visuals in anime, or remove that New York siren". It lists rollout from July 16, 2026 on Rapid Release domains and August 5, 2026 on Scheduled Release domains, and availability across Business, Enterprise, Education, Nonprofits and the Google AI Pro and Ultra consumer plans. It does not give resolution or length limits.
| Vids example | Prompt for gemini-omni-flash-1.1 edit | Status |
|---|---|---|
| Fix the color grading | Correct the color grade to neutral, natural skin tones. Keep everything else the same. | Visual edit |
| Restyle in anime | Restyle the visuals as hand-drawn anime. Keep everything else the same. | Visual edit |
| Remove a siren | Remove the siren sound. Keep everything else the same. | Audio edit, test first |
| Add a filmic look | Apply a soft filmic grade with lifted blacks. Keep the motion the same. | Visual edit |
The audio caveat
Google's own Omni announcement says speech and audio editing is still being tested before broader rollout. Vids lists removing noise, but Sume's docs do not describe an audio-only edit, so treat sound edits as an experiment on one clip, and use a deterministic tool for strict needs.
Run three looks over one clip
The loop below submits one edit job per look with its own Idempotency-Key, so a retry never double-bills.
import os, requests
LOOKS = {
"neutral": "Correct the color grade to neutral, natural skin tones.",
"anime": "Restyle the visuals as hand-drawn anime.",
"filmic": "Apply a soft filmic grade with lifted blacks.",
}
for name, text in LOOKS.items():
r = requests.post(
"https://api.sume.com/v1/video-router/generate",
json={"model": "gemini-omni-flash-1.1",
"prompt": text + " Keep everything else the same.",
"video_url": os.environ["SOURCE_CLIP_URL"], "resolution": "720p"},
headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
"Idempotency-Key": f"grade-{name}-001"},
timeout=60,
)
print(name, r.status_code, r.text[:120])
Reviewing a graded clip
Region rules are a legal question, not a technical one. If your users are in a region that Google excludes for Vids, ask your counsel before you assume that a different provider's route removes the question.
- Check skin tones in the first and last frames.
- Check that text on screen is still readable.
- Check that audio was not changed when you only asked for a visual edit.
- Keep the neutral version, so you can compare looks later.
Cost and the deterministic alternative
Three 8-second 720p looks cost $3.00 billed (3 x 8 x $0.10 x 1.25). If you only need a fixed grade, such as a vignette or grain, video filter applies a validated program of ops or an ffmpeg filtergraph on a Sume worker, with no model in the loop, so the result is repeatable frame for frame.
Sources
Related posts
More in Use cases
- GPT Image 2.5 21:9 first frame, then Seedance 2.5: a still for $0.01
A 2560x1088 GPT Image 2.5 still costs $0.0109 at medium quality and $0.0431 at high. Use it as the first frame of a 21:9 Seedance 2.5 clip on Sume.
- 48% thought Griffin was real: add a disclosure line to your clip
Tavus reports 48% of people took Griffin for a person after a one-minute call. If viewers cannot tell, label the clip: burn a disclosure line with caption cues.
- Guard a Sume run's text output with a yes/no model before you post it
Read output from a completed Sume run, ask a small decision model whether it is safe to publish, and treat a degraded outcome or null output as a hard stop.
- 300x600 half-page ad image: render 720x1440, keep a 600x1200 copy
A 300x600 half-page ad is 1:2. GPT Image 2.5 can render 720x1440 in one call; scale by 2.4 for 1x or by 1.2 for a 600x1200 retina file. Checks and prices.
Written by Sume