Gemini Omni 1.1 Flash Video Edit API by Google

Gemini Omni 1.1 Flash — Video Edit

Model ID: google/gemini-omni-1.1-flash/video-edit

Gemini Omni 1.1 Flash is Google DeepMind's natively multimodal model for video generation and editing. This variant takes an existing video plus a text instruction and applies the edit — adding, removing, replacing, or restyling elements — while preserving everything the prompt does not mention.

Overview

Gemini Omni Flash is a single transformer that processes text, images, audio, and video simultaneously rather than stitching separate modality-specific pipelines together. The result is output that is more cohesive, more consistent across turns, and more controllable than a chained generate-then-dub workflow.

What sets it apart from earlier video models is that it natively generates audio with every video — dialogue, ambience, music, and sound design are produced together with the picture rather than added afterward. The model is grounded in Gemini's real-world knowledge, so it reasons about physics, narrative logic, culture, and visual composition to produce results that feel intentional and cinematic. Every generated clip carries an imperceptible SynthID watermark alongside C2PA Content Credentials for provenance verification.

Because the model reads the whole clip rather than a frame at a time, an edit blends into the surrounding footage — lighting, motion, and untouched regions stay consistent instead of looking pasted on. Edits are also iterative: you can apply successive passes to the same material, and the model preserves your video across multiple amends.

What's new in 1.1

Version 1.1 is a control-and-fidelity upgrade over the first Omni Flash release:

• Flexible resolution output — new 4K and 1080p rendering for finished work, plus a fast 360p draft mode for previewing a shot before committing to a full-quality render.
• Longer continuous generation — shots can be extended segment by segment to a total of up to 40 seconds of coherent video.
• Precise shot start and end control — first and last frame can both be supplied, producing smoother camera moves, scene transitions, and seamless loops.
• Character and style consistency — a new video reference capability (clips of up to 3 seconds) substantially improves subject and art-direction stability.

How the five endpoints relate

AtlasCloud exposes Gemini Omni 1.1 Flash through five model IDs — text-to-video, image-to-video, reference-to-video, video-edit, and video-extend. All five route to the same underlying gemini-omni-1.1-flash model and differ only by the input modality they accept, corresponding to the model's task parameter (text_to_video, image_to_video, reference_to_video, edit, extend). This endpoint maps to edit.

Inputs

This variant takes a source video and a text instruction, with optional reference images.

• Prompt — natural-language description of the edit to apply (add, remove, restyle, or transform elements), up to 20,000 characters.
• video (required) — the source clip to edit. Must be a publicly accessible MP4 URL, 10 seconds or less.
• reference_images (optional) — 1 to 10 images used as character, scene, or style references for the edit. PNG, JPEG, JPG, or WebP, each up to 20 MB, supplied as public URLs or base64-encoded images.

Referencing images from inside the prompt

Instead of describing an uploaded image in words, point at it directly with <IMAGE_REF_N> — the Nth entry in reference_images, 0-based, so the first image is <IMAGE_REF_0>. Bindings may also be declared up front with a [# References <TAG>@MediaN] block at the start of the prompt, where the @MediaN handles are 1-based (@Image1 is the same item as <IMAGE_REF_0>).

Example: Replace the jacket the subject is wearing with the one shown in <IMAGE_REF_0>, matching the scene's lighting.

Key Capabilities

• Instruction-driven editing — add, remove, replace, or restyle elements of a clip from a plain-language description.
• Scene-consistent results — edits blend into the existing footage, preserving untouched regions, lighting, and motion.
• Reference-guided edits — supply up to 10 images to introduce a specific subject, object, or style, addressed inline with <IMAGE_REF_N>.
• Iterative refinement — apply successive edits to the same material without re-prompting the whole scene.
• Native audio — the accompanying soundtrack is regenerated in step with the picture where the edit calls for it.
• World-grounded realism — physics, motion, and scene dynamics informed by Gemini's real-world knowledge.
• Adjustable reasoning — the thinking_level control trades latency for quality on complex edits.
• Reproducible results — set a fixed seed to reproduce or iterate on a specific generation.

Input Parameters

Duration, aspect ratio, and resolution all follow the source video, so this endpoint has no duration, aspect_ratio, or resolution parameter.

Prompting Tips

• Name only what should change. Anything you leave unmentioned is preserved, so a narrow instruction produces a cleaner edit than a full re-description of the scene.
• Anchor the edit to the footage — "the car on the left," "the sign above the door" — so the model knows exactly what to touch.
• Chain small edits rather than asking for one sweeping transformation; each pass keeps the previous result intact.
• Use reference_images when the thing you want inserted is easier to show than to describe.
• Voice editing is not supported, and editing uploaded videos is unavailable in the EEA, Switzerland, and the UK.

Use Cases

• Element edits — add, remove, or replace objects, characters, or backgrounds in existing footage.
• Restyling — transform the look, colour grade, or mood of a clip while keeping its motion.
• Localization and cleanup — swap on-screen elements or refresh assets without reshooting.
• Reference-driven insertion — bring a specific product or character, supplied as images, into an existing scene.
• Iterative refinement — apply successive passes to converge on a desired result.

Pricing

Because this endpoint has no duration or resolution parameter, both are measured from the source video: its duration is parsed from the file, and its resolution is mapped to the nearest billing tier. Billing combines a per-second output charge at that tier, a per-reference-image charge, and a per-second source-video ingestion charge.

SKUs

Formula

video_duration = duration parsed from the source video
resolution_tier = tier probed from the source video

cost = min(10, max(3, video_duration)) × tier_rate
     + count(reference_images) × $0.00168
     + video_duration × $0.0087

Where tier_rate is $0.041 for the 144p/240p/360p tier, $0.11 for the 480p/720p tier, $0.16 for 1080p, and $0.31 for anything higher.

Notes and examples

• The output term is clamped to 3–10 seconds: a source shorter than 3 seconds is billed as 3, and the endpoint accepts at most 10 seconds of source anyway.
• The ingestion term uses the unclamped measured duration of the source video, at $0.0087 per second.
• Reference images are billed per item, not per second.
• An 8-second 720p source with 2 reference images costs 8 × $0.11 + 2 × $0.00168 + 8 × $0.0087 = $0.95296.
• A 10-second 720p source with no references costs 10 × $0.11 + 10 × $0.0087 = $1.187.
• A 10-second 1080p source with no references costs 10 × $0.16 + 10 × $0.0087 = $1.687.
• A 5-second 480p source with no references bills at the 720p tier: 5 × $0.11 + 5 × $0.0087 = $0.5935.
• A 10-second 4K source with 3 reference images costs 10 × $0.31 + 3 × $0.00168 + 10 × $0.0087 = $3.19204.