Gemini Omni 1.1 Flash for Music Videos: The Cut Between Two Frames Beats 4K

As of 28 August 2026, Google published Gemini Omni 1.1 Flash the day before. The three loudest words in the post are scene extension, 4K upscaling, and 360p drafts.

If you have actually finished an AI music video, you already know those three are not what kept you up. What usually breaks the clip is this: the chorus cuts, and the singer has a new face, a new jacket, and a new room.

The part of Omni 1.1 that can touch that failure is the least glamorous line in the announcement — first and last frame. This post does not retype the spec sheet. It answers what that control is for in an MV, and how to pick it tonight in SunoMV.

Split the announcement from the picker

Google’s write-up covers the full developer surface: the official post lists first-and-last-frame interpolation, 360p drafts (up to 60% faster and about one-third the cost of 720p), 1080p/4K upscaling, and scene extension — up to 10 seconds of prior context, in 10-second increments, to 40 seconds total.

The SunoMV video picker today exposes a shorter list:

Capability	SunoMV picker (Gemini Omni 1.1)
Inputs	Text, image, first and last frame
Duration	1–15 seconds
Resolution	720P or 1080P
Aspect	Landscape or portrait
Quota	250 credits / 5s at 720P; 1.5× at 1080P (375)
Not in this UI	4K upscaling, scene extension

Practical rule: Read the picker before you read the announcement. The announcement is what the model can grow into. The picker is what you can export tonight.

The product page with the full contrast lives at Gemini Omni 1.1 Flash. Below is only the music-video line.

The expensive part of an AI MV is the cut, not the pixels

A music video usually fails because two consecutive lyric lines do not look like they were shot on the same camera — not because a single frame is ugly.

Three failures you have probably shipped:

• New face: second 4 and second 12 are not the same person.
• New wardrobe: the chorus hits and the jacket changes colour.
• New room: the verse was a rainy bus stop; the chorus is another soundstage.

The old fix is editorial: generate unrelated clips, then force them together. A song is dozens of shots. Aligning them by hand costs more than generating them — which is why so many AI MVs collapse into a slideshow with transitions.

Practical rule: Judge an AI MV on the eight chorus seconds: how many cuts, and whether both sides of each cut are the same person.

First and last frame: pin the end of shot A as the start of shot B

First-and-last-frame on Omni 1.1 means you supply the opening picture and the closing picture. The model fills the motion in between.

For an MV that is a visual contract between two lyric lines. Export the last frame of shot A. Feed it as the first frame of shot B. Whip-pans, push-ins and orbits can go in the prompt — but the contract is locked first: person, wardrobe, camera side.

That is not the same move as writing “cinematic, 8K”. Empty adjectives lock nothing. Pinned frames do. We already unpacked why Veo prompting puts camera in the first slot in the Veo 3 prompt guide. Omni 1.1 turns the same idea into two stills.

Google’s own examples are camera contracts: a close-up that whip-pans into a saxophonist and a dancer, one continuous shot, no jump cuts. That is the chorus shot you actually need — not another pretty still.

We ran one: 360p first–last-frame, 60.4 seconds

The picker lists about 40 seconds of generation. Our own run on 27 August 2026, 360p first-and-last-frame: submit 5.4s, generation 57s, 60.4s end-to-end. The sample clip is here.

That is not a vendor SLA. It is one measured run. The only point: the eight chorus seconds you need tonight come back in minutes, not hours.

Quota is on the picker too: 250 credits per 5 seconds at 720P, 375 at 1080P. Board the song at 720P. Restage two chorus shots at 1080P. Do not spend the HD budget on a draft that has not locked a face yet.

The cheaper iteration row is MiniMax H3 Max: 480P/720P, 5–15 seconds. Storyboard there, keepers here. Two rows on one line, not rivals.

Practical rule: Use 720P to learn whether the cut holds. Spend 1080P only on the shots that already hold.

When Omni 1.1 is the right row — and when it is not

Use it when you already have two lyric stills (or two pictures that must connect) and you need a real camera move, not a dissolve. Chorus, hook, the cut a viewer will rewind.

Skip it when you are still hunting a visual theme — board twenty cheap shots first. When you need native 2K, lip-sync and multi-reference lock — that is another row, not Omni 1.1. When you want 4K upscaling or a 40-second scene continuation — the announcement has it; the SunoMV transition UI does not, as of this post.

If you already have a Suno track, start from Suno to Video and switch the video model to Gemini Omni 1.1. You do not leave the music-video flow to apply for a developer key.

Pick it today

1. Open SunoMV.
2. Paste the song (a Suno link or your own file).
3. Choose Gemini Omni 1.1 in the video-model list.
4. Set 1–15 seconds, 720P or 1080P, landscape or portrait.
5. Pin first and last frame when two lines must connect; otherwise use text or one still.
6. Stack on the lyric timeline, preview free, export.

Specs, quota, the measured run, and the “announcement vs picker” contrast are on Gemini Omni 1.1 Flash. This post only has to do one job: do not get pulled around by 4K. Land the chorus cut on the same person.