Omni 1.1 first and last frame

Omni 1.1 Flash first-and-last-frame is interpolation between two stills you supply, not a prompt that invents an ending. Google’s 27 August 2026 Keyword post says you specify start and end frames and Omni 1.1 generates continuous video between those keyframes — orbits, zooms, transitions, loops. That control lives on gemini-omni-1.1-flash in AI Studio, Agent Platform, and Flow. It is not the Versely slug VEO First Last Frame, and it is not the definition page at first-last-frame. Three names. One job shape. Three meters. Versely does not publish /models/gemini-omni-1-1-flash.

This page is only Omni 1.1’s two-still interpolation versus those other two doors. Scene extend is time. This is space between two pictures.

Two stills, or you are not in this control

First-last is a bounded problem. Ordinary image-to-video knows where it begins and improvises the rest. First-last also knows where it must land. The model is contracted to open on still A and finish on still B. If B does not exist, there is nothing to land on, and you are in a different job: animate this plate.

Google’s 1.1 examples are camera sentences between two pictures: a low-angle drummer that whip-pans onto a saxophonist and a dancer; a zoom into a TV that returns to the opening scene as a loop. The prompt describes how to travel. The stills describe where. Rewrite the prompt all day; if the end still is a different room, a different lens family, and a different person, the travel will read as a morph. That is the glossary pitfall, and it applies here.

What to lock in the two stills before you generate:

• Same location, or a travel that a camera could actually do in the duration you will get.
• Same light family. Daylight A to tungsten B in four seconds is a grade, not a move.
• Same subject identity. First-last is not reference-to-video. It will not recast a new actor into still B while keeping still A’s face. If you needed recast, that is VEO 3.1 Reference to Video or Omni Video’s reference path, not this control.
• A composition that can interpolate. A tight product hero to a city aerial is a warp.

Loops are the special case: last frame identical to first, motion in the middle. Google calls out seamless looping clips. Identical files in both slots are the loop contract. Near-identical files are a jump cut wearing a loop label.

You still write a prompt. “One continuous shot, no jump cuts” is in Google’s own examples for a reason. The stills bound the ends. The sentence bounds the path.

On Versely, if you needed this job on a published Google row, you open VEO First Last Frame: image-to-video category, requires an image, 4s / 6s / 8s, 4K, from 64 credits, native audio on. That is Veo interpolating two frames, with Veo’s audio and Veo’s length menu. It is the closest catalog cousin. It is not Omni 1.1.

How this is not the glossary page, and not Veo

The glossary page defines the method. First-last-frame generation takes two stills and generates the motion between them. Loops and transitions are the easy wins. Unrelated compositions are the artefact everyone notices. Read first-last-frame when you need the term. Do not treat the term as a SKU. A definition does not bill credits and does not pick Omni versus Veo.

Veo first-last is the catalog row. Audio is on. Durations are 4, 6, or 8 seconds. Resolution lists 4K. Credits start at 64. The family is Veo 3.1: cinematic physics, always-on dialogue when you ask for it, SynthID in Google’s stack. You use it when the interpolation should look and sound like Veo, and when you want that job on the same bill as Veo 3.1 generate.

Omni 1.1 first-last is the 27 August control on gemini-omni-1.1-flash. Google’s pitch is full-context reasoning for camera moves that read as real — arcs, pushes, pulls, whip-pans — between two keyframes. Access is AI Studio, Agent Platform, Flow for Plus / Pro / Ultra. Duration, audio flag, and credit price for that control are Google’s meter, not a Versely card. Do not copy Veo’s 64 credits onto it. Do not copy Omni Video’s 29 credits onto it.

Gemini Omni Video can take images as references on a 4–10 second generate (quota images + videos*2 + character_ids <= 7). That is multimodal generate, not a first-last interpolator. Feeding two images into Omni Video as “refs” does not contract the model to open on file A and land on file B. Reference is casting. First-last is endpoints.

Gemini Omni Flash Edit is video-to-video from 50 credits. It changes a take you already have. It does not interpolate two stills into a new take.

If a teammate says “just do first-last,” ask which of the three they mean. If they mean the method, any interpolating row. If they mean the Veo card, 64 credits and 4/6/8s. If they mean Omni 1.1’s new control, Google’s door that day.

Briefing the path, not the vibe

The prompt is the camera. The stills are the set.

Good path language: slow orbit keeping the bottle center; whip-pan from the drummer to the sax; push into the TV until the inner scene fills the frame; locked-off dissolve is not a camera, say so if you wanted a morph.

Bad path language: “epic cinematic transition, emotional, award-winning.” First-last will still connect A to B. The connection will be whatever the model uses when you declined to steer.

Duration must be plausible. Google’s 1.1 post does not reprint a Versely duration menu for this control. On Veo first-last you pick 4, 6, or 8 seconds. A 4-second travel from a wide kitchen to a macro of a label is a snap zoom, not a tour. If the move needs twelve seconds, first-last on Veo cannot hold it in one pass. Omni 1.1 scene extend is a different control (ten-second memory, forty-second cap) and does not replace a missing end still.

Shoot or generate the stills as a pair. Same camera height, same grade, same product angle unless the move is the point. Flux 2 Max is a 6-credit stills row if you need to invent B before you interpolate. Do not ask first-last to invent B in prose. That wastes an interpolation on a destination that never existed.

Audio: Veo first-last has audio: true. Write the room or you will get a room you did not ask for. Omni 1.1’s demos include score and speech in other controls; do not assume a mute interpolation. If you needed a silent plate for a later voice, say so in the prompt, then finish with captions in the editor. Add captions to video after the file exists.

Article 50 still applies to a realistic synthetic. Two stills of a real person plus an interpolated orbit is still generated video. Label it when you are the deployer. Template: Article 50 disclosure.

When to refuse first-last

Refuse Omni 1.1 first-last when you only have one still. That is image-to-video. On Versely, image-to-video is the door. Omni Video accepts a still. Hailuo 2.3 Fast requires one. First-last requires two.

Refuse it when the job is identity in a new scene. That is reference-to-video. Veo R2V is 8s, from 128 credits, up to three stills, native audio. The stills are a cast, not endpoints.

Refuse it when the job is “keep this Omni scene going.” That is 1.1 scene extend: ten seconds of prior video context, ten-second steps, forty-second cap. You need a clip, not two stills.

Refuse it when the job is a 30-second continuous action with no endpoints. That is Seedance 2.5, 4–30s, 720p, from 34 credits (was 67).

Use it when you already approved two frames and the content is the travel: product from closed to open, chair from empty to occupied, camera around a locked hero, a loop for a sticker. Then pick the meter: Veo first-last on this bill, Omni 1.1 first-last on Google’s 1.1 surface.

The AI video generator runs the published rows. For the Veo interpolator, open that card. For Omni 1.1’s version, use Google’s door and bring the file back.

If you want the Google Omni rows we actually bill, sign up and run Omni Video or Omni Flash Edit — and use Veo first-last when two stills are the brief on this catalog.

FAQ

Is Omni 1.1 first-last the same as /glossary/first-last-frame?

No. The glossary defines the method: two stills, motion between them. Omni 1.1 is Google’s 27 August implementation of that method on gemini-omni-1.1-flash. The glossary does not generate video.

Is it the same as VEO First Last Frame?

No. VEO First Last Frame is a published Veo row: 4 / 6 / 8s, 4K, from 64 credits, native audio. Omni 1.1 first-last is a 1.1 control on Google’s surface. Same idea, different family and meter.

Can I fake the last frame with a prompt?

You can try. You will pay interpolation prices for image-to-video behavior. Generate still B first — Flux 2 Max is a stills row — then interpolate. One still is a different job.