Generate and Edit Videos with Gemini Omni Flash

A skill for generative video editing, text-to-video, image-referenced video generation, and first-frame-to-video transitions using the Gemini Omni Flash model (gemini-omni-flash-preview) via the official google-genai SDK, plus workflows for pre-processing high-resolution or long source videos with ffmpeg. Four core capabilities: video editing and refinement (max 10-second clips, stylistic changes, inpainting/outpainting), text-to-video generation from a prompt, first-frame-to-video generation from a single input image, and image-referenced generation using style/character/object references. Video uploads for video-to-video edits are unavailable in the EEA, Switzerland, the UK, and some US states - a fast completion with total_output_tokens: 0 or empty video output is the tell for hitting this restriction. All operations go through the Interactions API; input media must first be uploaded via the Files API, with the returned file URI and MIME type included in the interactions.create input parts array. Requires google-genai >= 2.10.0, Python >= 3.10, and ffmpeg/ffprobe on PATH for prep_video.py, inspect_video.py, and generate_video.py's --strip-audio option. Three bundled scripts drive the workflow: upload_file.py uploads local media to the Files API and polls until ACTIVE, warning if a video exceeds 25MB (Omni Flash is optimized for 10s clips at 720p/24fps) and recommending prep_video.py first; generate_video.py performs end-to-end generation supporting text-to-video, image-to-video (single reference or two-image interpolation between keyframes), video editing with original audio kept or fully regenerated via --strip-audio, turn-by-turn editing of a prior generation via --previous-interaction-id (no re-upload needed), and parallel batch execution from either a line-by-line prompts file or a fully-configured JSON jobs array with per-job prompt/video/image/output/strip_audio/aspect_ratio settings; and inspect_video.py checks a local video's duration, resolution, FPS, audio presence, and format via ffprobe, with a --json flag for structured output. Generation is configured with --aspect-ratio (e.g. 16:9, 9:16) and --duration (any integer 3-10 seconds), and outputs are saved locally (e.g. under media/).

Use it for generative video tasks with Gemini Omni Flash - editing existing short videos, generating video from text or a reference image, or interpolating between two keyframe images. Video-to-video editing (uploading a source video) is unavailable in the EEA, Switzerland, the UK, and some US states.

Input is a text prompt plus optional reference image(s) or a source video (uploaded via the Files API first), with aspect ratio and duration settings. Output is a generated or edited MP4 video saved to the local filesystem, or structured video metadata (duration, resolution, FPS, audio) from inspect_video.py.

pip install -U google-genai

Requires the Interactions API and Files API from the Gemini API, ffmpeg/ffprobe for video pre-processing and inspection, and pairs with the broader Gemini Interactions API skill for platform-wide model/SDK rules.

Developers building generative video features - text-to-video, image-to-video, style transformation, or keyframe interpolation - who need the exact script interface, regional restriction awareness, and file-size/format guidance for working with Gemini Omni Flash reliably.