Gemini Omni 1.1 Flash First to Last Frame Video Generator

Last verified: November 1, 2026

Gemini Omni 1.1 Flash works best on shots where the start and the finish are already decided: a camera circling a subject, a zoom from a wide view into a detail, or a clip that loops back to where it began. It delivers these shots at up to 4K. Google built the model's start-and-end-frame control for these moves, and the model generates continuous video between two keyframes. Gemini Omni 1.1 Flash First to Last Frame puts that control into a short form: you upload the opening still and the closing still, describe the move, and the model creates the motion between them.

Google documents this workflow as interpolation: the model animates the scene from the first frame to the last. In practice that makes it a storyboarding tool as much as a generator. You settle the two key compositions as still images, which are quick to retouch or reshoot, and leave the motion in between to the model. Then you pick the length, resolution and frame shape to suit where the clip will play.

Explore more first to last frame

First to Last Frame Use in Text To Video Use in Image To Video Use in Reference To Video Kling 3.0 Kling O1 Pro Seedance 2.5 MiniMax H3 Max Kling 4.0 Wan 3.0 Prime
Capability Snapshot

Key Specs for Two-Frame Generation

What goes in and what comes out, at a glance.

Input

First Frame (JPG, JPEG, PNG or WEBP, up to 10 MB) + Last Frame (JPG, JPEG, PNG or WEBP, up to 10 MB) + a text prompt of up to 20,000 characters

Video Duration

4, 6, 8 or 10 seconds

Resolution

360P, 720P, 1080P or 4K

Aspect Ratio

9:16, 16:9

Output

1 video per run

How Do You Get a Smooth Transition Between Two Frames?

Spend the Prompt on the Motion, Not the Look

Both stills already show what the scene looks like, so use the Prompt box to describe the path between them. Google's own example for this workflow is one line about change, a smooth timelapse from sunrise to sunset, not a description of either image. Name what moves (the subject, the camera, the light) and which way it goes. Add appearance details only for things neither frame shows.

Pin Each Beat to a Moment in the Clip

Google's prompt guide says you can ask for things to happen at specific times in plain language, with no special syntax. Use that to pace the transition against the Video Duration you choose. Hold the opening composition briefly, say when the move starts, and say that it settles on the closing framing at the very end. That way the arrival at your last frame is planned instead of squeezed into the last moment.

Draft Small, Then Re-Render the Chosen Take Large

Google positions the smallest resolution as a lightweight preview that generates up to 60% faster than the model's standard resolution, and suggests upscaling only finished projects to 4K. One of its example apps makes three or four drafts that each change one thing, then compares them side by side. Work the same way here: set Resolution to its lowest option while you test frames and prompt wording. Then rerun the chosen frames and prompt, unchanged, at the top option for delivery. Judge that final run on its own, because it is a new generation.

How Do You Turn Two Frames Into One Video?

Six steps take you from two stills to a downloaded clip.

1

Add your opening image in "First Frame"

Click "Upload first frame" and choose the still the clip must open on. The model animates away from this exact composition, so place the subject where you want the move to begin, and prepare the image in the orientation you plan to export.

2

Add your closing image in "Last Frame"

Click "Upload last frame" and choose the still the clip must arrive at. Both slots are required. Prepare this image in the same orientation and at a similar quality as the first, so the two ends of the clip match.

3

Describe the move in the "Prompt" box

The "Prompt" box is required and holds up to 20,000 characters, far more than a transition needs. Describe how the subject, camera and light travel from the first image to the second. If you only have a rough idea, press "Enhance Prompt" under the box for fuller suggestions. Your text changes only if you pick one of them.

4

Choose "Video Duration", "Resolution" and "Aspect Ratio"

"Video Duration" offers 4, 6, 8 or 10 seconds and starts at 4. Pick the length that fits where the clip will be used. "Resolution" offers 360P, 720P, 1080P or 4K, with 720P preselected. "Aspect Ratio" defaults to 9:16, and 16:9 is there for landscape frames. The optional "Seed" field takes a number. Keep the same number in it while you revise the prompt, so the prompt is the only thing you changed.

5

Click "Generate"

Before you click "Generate", check that each image is in the right slot. The model animates from the first frame toward the last, so swapping the two images reverses the direction of the move.

6

Download the result

Each run returns one video. Watch it to the final second to confirm it arrives at your closing image before you download it. If the path between the frames is wrong, revise the prompt before you replace the frames.

Choosing Between Two Stills and Existing Footage

When to pin both ends of a shot, and when to start from video you already have.

What you start from You bring two still images, one setting the opening composition and one setting the closing composition, plus a prompt for the motion that joins them. The Edit Video mode of the same model starts from video you already have, which the model uses as the source to modify according to your written instruction. Pick the two-frame workflow when the clip must land on a specific composition, and pick Edit Video when the footage already exists and only needs to change.
Kind of job it suits It suits shots defined by where they begin and end, such as a camera circling a subject or a push from a wide view into a close detail. Editing suits footage whose motion already works, and Google's own examples include restyling a clip as anime and changing the text on a sign. Choose two frames when the camera move itself is the shot, and choose editing when the movement is right and only the look or a detail needs to change.
How the prompt should read The prompt describes the journey between the frames, in the spirit of Google's own one-line example of a smooth timelapse from sunrise to sunset. Google's prompt guide says simple prompts work best for video editing, so an edit is best written as a short instruction about what to change. Writers who think in camera paths will feel at home with the two-frame prompt, while a single quick change to existing footage is better written as a short editing instruction.
Use Gemini Omni 1.1 Flash with two frames when the opening and closing compositions are fixed and the move between them is the point. Switch to Edit Video when you already have footage and want to change it rather than create the motion.

Fixing Cuts, Warping and Rushed Transitions

The problems that come up most with two-frame clips, and what to change before the next run.

The video cuts to another shot instead of moving straight from my first frame to my last

By default the model tries to build a video out of a few different shots, and Google's prompt guide says that a single scene has to be asked for. Start the Prompt with an explicit instruction such as 'one continuous shot, no cuts', keep both frames as they are, and generate again.

The subject warps or melts halfway through the transition

Google's model card lists scenes with complex motion as a remaining challenge, and a big jump between two very different frames asks for exactly that. Bring the frames closer together, with a similar pose and a similar camera angle, or reduce the prompt to one action and one camera move. Then run it again.

The move feels rushed and only just finishes as the clip ends

Video Duration starts at its shortest length, which leaves little room for an orbit or a long push. Raise it to one of the longer options and generate again. The frames and prompt can stay the same.

My image won't upload to First Frame or Last Frame

Each slot accepts only a few still-image formats, and the file must be under a size cap and above a minimum pixel size. All of these are listed in the FAQ on uploads. Convert the file to an accepted format, export a lighter version if it is too large, or use a bigger original if it is too small, then upload it again.

Frequently Asked Questions

What images can I upload as the first and last frame?

Both slots are required, and each takes one still image in JPG, JPEG, PNG or WEBP format, up to 10 MB and at least 300×300 pixels. The minimum is a floor, not a target, so upload the cleanest, largest version of each image you have within the size cap.

Should my frames be vertical or horizontal?

Set Aspect Ratio to match the orientation of your two stills. The form starts on vertical, so switch to widescreen before generating if your frames are landscape. Prepare both images in the same orientation so the opening and closing compositions line up.

Is the 4K option native 4K?

Google presents the model's 4K as upscaling: its launch post describes crisp 4K upscaling and suggests upscaling finished projects to 4K for a polished look. Choose it in Resolution when your delivery format calls for that size.

What makes one generation cost more than another?

Once you're signed in, the form calculates the credit cost of a run from the options you've picked and shows it above the Generate button. The figure changes when you change Video Duration or Resolution, so check it again after adjusting either one.

What frame rate does Gemini Omni 1.1 Flash produce?

Google lists the model's video output at 24 frames per second. Plan your edit around that rate, and check the frame rate of the downloaded file before you add it to a timeline.

References

Sources and citations used to support the content provided above.

Updated: 2026-11-01 13:37:46 3 Sources

blog.google

Source Link
https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/

ai.google.dev

Source Link
https://ai.google.dev/gemini-api/docs/omni

aistudio.google.com

Source Link
https://aistudio.google.com/docs/omni