Superside Gemini Omni Flash Edit

Video editing without a timeline: connect a video, type what you want changed, get a new video. That's this node - Google's Gemini Omni Flash running on fal, taking a video and a text instruction and returning the edited clip. "Make this video anime. Keep everything else the same." is the kind of prompt it's built for: one clear change, everything else preserved.

The wiring is ComfyUI-native, which is the nicest thing about it. There's no manual URL copying - you connect a LoadVideo node straight into the video input, the node uploads the file to fal internally, and the edited result comes back as a VIDEO that plugs directly into SaveVideo or PreviewVideo. Inputs are just video, prompt, and api_key. Outputs are video (VIDEO) and video_url (STRING, a direct fal link in case you want to grab the file elsewhere).

The prompt guidance is worth following literally. Keep prompts short and add "keep everything else the same" - Omni Flash is an edit model, and like every model in its class it will happily reinterpret parts you didn't mention unless you tell it not to. That's the same drift failure the KB documents across the instruction-edit family: whole-frame regeneration comes back close-but-not-identical, and it compounds if you chain edits. One clean edit per pass is the reliable pattern.

Two things to check before you build a pipeline around this. First, there are regional restrictions: the README and node are explicit that editing uploaded videos isn't available in the EEA, Switzerland, or the UK, and voice editing and audio references aren't supported at all. If you're in one of those regions, the node will fail on you no matter how good your prompt is. Second, it uses fal's queued execution path - the pack's base class routes Gemini Omni Flash through fal's submit-and-poll queue with a 20-minute client-side timeout, because a single long-held connection on a multi-minute generation is prone to mid-flight disconnects. Practically: the node can take a while, the graph waits (with a proper timeout rather than hanging forever), and you shouldn't panic if it doesn't return instantly.

The honest framing, same as every node in this pack: it's a fal API call, metered credits, your video leaves your machine, and the model's content policy travels with it. There is no local version of Gemini Omni Flash - this is a closed model, and this node is the only door into a Comfy graph for it. If you want offline video-to-video with this shape of editing, that's a different (open) model entirely, like Wan.

Install - ComfyUI Manager (search "comfyui-superside-nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI, find it under Superside. Paste your fal key into api_key (blank falls back to FAL_KEY, both blank errors immediately). No model downloads - the only real dependency the pack adds is fal-client.