Generate and conversationally edit short videos with Google Gemini Omni Flash.
npx skills add https://github.com/calesthio/OpenMontage --skill gemini-omni
Please help me install this Agent Skill. Skill: gemini-omni Source: https://github.com/calesthio/OpenMontage/tree/main/.agents/skills/gemini-omni Command: npx skills add https://github.com/calesthio/OpenMontage --skill gemini-omni
Iterating on AI-generated video normally means regenerating the entire clip from scratch, losing everything that already worked. This Skill wraps Google Gemini Omni Flash so you can generate 3-10 second 720p clips with synthesized audio and then refine them conversationally — editing only what you describe while everything else stays the same.
interaction_id; pass it back as previous_interaction_id with a short edit prompt (e.g., "Make the phone invisible. Keep everything else the same.") to refine the same clip in layers.<FIRST_FRAME> and <IMAGE_REF_N> tags for consistent characters, products, or styles.[0-3s]-style timecodes and render timed on-screen text, with automatically synthesized audio directed via the prompt.Ask the agent to generate a 6-second 720p clip of your scene using gemini_omni_video, then follow up with a short edit instruction referencing the returned interaction ID to refine it.
High-intent search queries and answers about installing and using this skill.
Generate the base clip with gemini_omni_video, which returns an interaction_id. Pass that ID back as previous_interaction_id with operation edit_video and a short prompt describing only the change, ending with "Keep everything else the same."
Pass local images via reference_image_paths, then bind them to roles inside the prompt using <IMAGE_REF_N> tags indexed from 0 in the order supplied. Use <FIRST_FRAME> to make an image the opening frame of the clip.
Output is limited to 3-10 second clips at 720p and 24fps in MP4 format with audio, in 16:9 or 9:16 aspect ratio. There is no seed, negative prompt, temperature, or video extension support, and all output carries a SynthID watermark.
Yes, pass input_video_path instead of previous_interaction_id and the tool uploads the video via the Files API for editing. This is unavailable in the EEA, Switzerland, and the UK, though editing generated videos works everywhere.
Editing via previous_interaction_id only works if the prior call stored the interaction server-side, which is the default. If store was set to false on the original generation, the interaction state is gone and the clip cannot be edited.
Use Seedance 2.0 for one-shot cinematic hero clips and veo_video for first/last-frame interpolation. Gemini Omni is the right choice for iterative conversational editing, on-screen text, and reference-image-bound subjects in clips under 10 seconds.
anxiety135790Convert audio, video, images, or text into colored ASCII video with MP4, GIF, or frame-sequence output.
gooseworks-aiConvert an EDL into a finished master.mp4 video using ffmpeg.
kagentiRecord narrated demo videos of the Kagenti platform using Playwright and OpenAI TTS.
gyalamanch001aGenerates AI videos from text prompts and reference images via selectable providers.