Gemini Omni 1.1 Flash Edit Video: Change Clips by Prompt

Last verified: October 29, 2026

Gemini Omni 1.1 Flash works best for targeted changes to short footage you already like. You can give the whole take a new look, add something to the scene or make something disappear, and you ask for each change in plain language. Google built the model for high-speed video generation, editing and cinematic control. Its developer guide sets up an edit of an uploaded clip as a video plus a text instruction. Every Gemini Omni 1.1 Flash edit video run here works the same way: your clip goes in, your sentence describes the change, and an edited video comes back.

For a working creator, this means you can answer a note on a shot (wrong prop, wrong mood, wrong style) with a sentence instead of a reshoot or a rebuild. The model is built to follow both simple and complex instructions and to edit video through conversation. On the form, each run applies one instruction to the clip you upload in Video. Add Reference Image holds any pictures the change depends on, Prompt holds the instruction, and Resolution sets the size of the file you get back.

Explore more edit video

Use in Text To Video Use in Image To Video Use in Reference To Video Use in First To Last Frame Kling 4.0 Seedance 2.5
Capability Snapshot

Inputs and Output at a Glance

What the edit form accepts and what each run gives you back.

Input

Video (MP4 or MOV, up to 50 MB) + up to 5 images (JPG, JPEG, PNG or WEBP, up to 10 MB each) (optional) + a text prompt of up to 5,000 characters

Resolution

360p, 720p, 1080p or 4K

Output

1 video per run

Prompting Techniques for Cleaner Edits

Cut the Instruction Down to One Action

Simple editing prompts are the documented way to brief this model. Google's own before-and-after examples replace a long sentence about where something enters, how it moves and what happens next with a short version that names the action and stops. Write your edits the same way: one verb, one object, one result, for example Give him a red scarf. Don't fill the prompt with a description of the scene your clip already shows.

End With a Line That Protects the Rest of the Shot

When only one element should change, Google recommends ending the instruction with "Keep everything else the same" to keep the rest of the video consistent. Use it as a standard ending for any targeted edit, like a wardrobe swap, a lighting tweak or a removed object. Leave it out only when you want the whole clip reworked, as in a full style change.

Tie an Effect to a Moment in the Action

Google's example edit for an uploaded clip doesn't just ask for an effect. It says when the effect happens: when the person touches a mirror, the glass ripples like liquid and their arm turns into mirror material. Write your own effects the same way, starting from an event that's clearly visible in your footage, like a hand landing on a table or a door swinging open. That way the instruction names both the trigger and the result.

How to Edit a Video in Six Steps

Six steps take you from uploading a clip to downloading the edited version.

1

Upload your clip in Video

Use the "Click to upload the video to edit" area in "Video" to add your footage. It's the only required upload. Trim the clip to the part that matters before you upload it, because the form only takes short clips. Pick a take where the element you want to change is clearly visible and not blocked by anything else in the frame.

2

Add optional images in Add Reference Image

Skip "Add Reference Image" when words can carry the edit, as with a style or lighting change. Add pictures when the change depends on something specific, like an exact location, product or outfit that should appear. Plan to say in your prompt what each image is for.

3

Describe the change in Prompt

Type the edit into "Prompt". The box holds up to 5,000 characters, but a targeted edit rarely needs more than a sentence. State the one change and describe any reference image by what it provides. If the rest of the shot should stay as it is, add the closing line covered in the techniques section.

4

Pick an output size in Resolution

"Resolution" offers 360p, 720p, 1080p and 4K, and 720p is selected by default. Use 360p to check quickly whether an instruction works. Keep 720p for everyday review copies, and save 1080p or 4K for the version you'll deliver. A re-run at a higher setting is a new generation, so check that result again too.

5

Start the run with Generate

Click "Generate" once the clip, any images and the prompt are in place. Each run returns one edited video. To compare two wordings, run them one after the other and note what you changed between them.

6

Download the result

Download the edited video and watch it at full size before you use it. First check the element you asked to change, then scan the rest of the frame for anything that shifted. If something is off, adjust the prompt and run it again.

Fixing Common Editing Problems

What tends to go wrong with prompt-based edits, why it happens and what to change before the next run.

The edit changed parts of the shot I never asked about

Google's model card says that keeping complete consistency throughout edits is still a challenge for the model. Its editing guide addresses this with short prompts and a closing instruction to keep the rest of the video unchanged. Rewrite your prompt so it names only the element to change, add the closing line from the techniques above, and run it again. If parts of the shot still shift, try a take where the element you're changing is larger and easier to pick out in the frame.

The new text on a sign or label is misspelled or garbled

Changing the text on a sign is one of the simple edits Google documents, but its model card also lists perfectly accurate text rendering as an open challenge. Keep the new wording short, put it in quotation marks inside the prompt as Google's example does, ask for nothing else in the same run, and regenerate if a letter comes out wrong. For copy that must be exact, like a brand name or a legal line, add it in your editing software afterwards.

Fast movement in my clip looks warped after the edit

Google's model card lists complex motion among the challenges the model still faces, so clips with fast, tangled action are the riskiest to edit. Choose a take with calmer movement that's easy to follow, or aim the edit at something that stays still in the frame instead of a fast-moving limb or object. Then run it again.

My video won't upload

The Video field only takes short clips in two formats and under a file-size cap, all listed in the upload question below. A longer recording, a different file type or a heavy high-bitrate export will be refused. Trim the clip to the part you want edited, export it in one of the accepted formats, lower the bitrate if the file is still too big, and upload it again.

My reference image is rejected

Add Reference Image checks each picture's format, file size and minimum dimensions, and limits how many you can attach. The limits are listed in the upload question below. Re-save oversized images at a smaller file size, swap small thumbnails for the original photo, convert other file types to one the form accepts, and keep only the images your prompt actually refers to.

Frequently Asked Questions

What video and image files can I upload, and how long can the clip be?

The Video upload is required and takes an MP4 or MOV file of up to 50 MB and no more than 10 seconds. Add Reference Image is optional and accepts up to five pictures in JPG, JPEG, PNG or WEBP format. Each picture can be up to 10 MB and must be at least 300×300 pixels. If your footage is longer, cut out the part you want changed before uploading.

What makes one edit cost more than another?

The credit cost of each run is calculated from the options you choose and shown above the Generate button once you're signed in. Resolution is the setting that changes it, so check the cost again after switching to a different output size.

Is the 4K output from Gemini Omni 1.1 Flash native or upscaled?

Google presents 4K as upscaling. Its announcement lists 1080p and 4K under the model's upscaling feature and describes them as polished outputs ready for professional production. In the Resolution menu, choose one of those two options when the file has to hold up on a large screen.

Can Gemini Omni 1.1 Flash change the camera angle of existing footage?

Yes. Google DeepMind lists camera angles, along with details and environments, as things you can change when editing. Its multi-turn demo includes an instruction that switches a violinist clip to an over-the-shoulder view. Ask for the new viewpoint in one sentence and save other changes for a separate run, so you can judge the new angle on its own.

How do I tell Gemini Omni 1.1 Flash what my reference images are for?

Say it in the prompt. Google's prompt guide recommends a plain closing instruction stating that the images are references, not literal starting frames. Its examples describe each image by the role it plays, such as the violin a performer holds or the woman who walks through the shot. For an edit, describe what the picture provides (the location, the outfit, the product) and what in your clip it should change.

References

Sources and citations used to support the content provided above.

Updated: 2026-10-29 21:19:49 3 Sources

ai.google.dev

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

deepmind.google

Source Link
https://deepmind.google/models/model-cards/gemini-omni-flash/

deepmind.google

Source Link
https://deepmind.google/models/gemini-omni/