Gemini Omni

Google's video generation has been the closed benchmark since Veo's native audio first shipped, and this node is its entry in Civitai's pack under the "Gemini Omni" brand. Two things make it stand out from every other video node here: the model selector and the mandatory max_billable_seconds input. The second one is a rare bit of self-restraint in an API wrapper - the node makes you declare, up front, how many seconds you're willing to be charged for, and that cap is a required field, not a courtesy toggle. When a closed video model is metered per second, that's a genuinely useful tripwire.

It's part of civitai-comfy-nodes, Civitai's official ~160-node pack generated from its Orchestration API spec. Cloud job like the rest: submit a videoGen tagged gemini-omni, long-poll, get a native VIDEO tensor. Billed in Buzz, nothing local.

The inputs that matter

• prompt (required, multiline) - the whole creative direction.
• max_billable_seconds (required INT) - your spending ceiling in output seconds. This is the "no surprise bill" knob; set it to what you'd actually pay for and let the job fail if it would exceed it. If it's new to you: yes, a 0 here likely means the job won't go anywhere, so set a real number.
• model - an enum that's empty by default with 1.1-flash available. Empty = the service's default choice; pick 1.1-flash for the lighter, faster variant.
• aspect_ratio - 16:9 or 9:16. No landscape/portrait menu sprawl here.
• resolution - 360p, 720p, 1080p, 4k - and an empty default meaning "let it decide." 4k is on the table, which most of the other engines in this pack can't even offer.
• images - optional IMAGE socket for image-to-video when you have a still to start from.

Outputs: video (VIDEO), draft_cache_audio (AUDIO), draft_cache_video (VIDEO), workflow_id and raw_json (STRING).

Install and auth

Standard pack install - official, Comfy Registry:

# ComfyUI Manager: search "Civitai Comfy Nodes"
cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

Restart, authenticate via a Civitai Auth node, CIVITAI_API_TOKEN env var, or sidebar OAuth.

Common issues

The 4k option is a trap for the unwary spender. 4k per-second billing on a long clip is where a "let me just try it" session becomes a real invoice; keep resolution modest until you've confirmed the motion works, then promote. max_billable_seconds is easy to misunderstand - it's your cap, and a job that would exceed it fails rather than overcharging, which is the behavior you want, but it means a hard-failing job isn't necessarily an error on your end. Your prompt and images leave your machine for Google's servers. And the pack is early preview - nodes and behavior can change without notice. Google's models also carry the vendor's own content moderation, which you can't patch around (there are no weights to edit) - if a prompt refuses, that's the model's policy, not the node's bug.