Google Gemini Omni Flash Text To Video (Channel Edition)

Text in, short video out. This is the text-to-video member of the Gemini Omni Flash pair that ships in BizyAirPlus, and it's the most hands-off way to get a moving image inside ComfyUI. You type a prompt, pick a resolution and a duration, and a clip comes back. No checkpoint download, no VRAM, no sampler settings to argue with.

It belongs to the BizyTRD model-service family - cloud nodes that call hosted models through SiliconFlow's BizyAir platform. The whole pack is built around a switch in your action bar: flip it ON and supported nodes run on BizyAir's servers, flip it OFF and everything runs locally as usual. This node only does anything when the switch is on.

Why you'd reach for it

If you've been generating stills locally and want to dip a toe into video, this is about the lowest-friction entry point there is. No installing Wan, no hunting for a 20GB checkpoint, no learning what a sigma schedule is. Just a prompt and a couple of dropdowns. The flip side is that it's a closed, paid API - you're renting the generation, and the community's running debate about closed video models applies fully here. For the price of a few seconds of billed compute you get Google's omni model doing the thing you'd otherwise spend an evening setting up locally.

How it works

The inputs are refreshingly few. prompt is the whole game - 1–2048 characters, and the tooltip is explicit about what belongs in it: scene, characters, action, environment, camera movement, screen style, and audio cues. Gemini Omni generates audio natively, which is the one capability open video models genuinely still lack, so mentioning sounds and voice in the prompt isn't wasted effort.

Then resolution (720p by default) and duration (default 4) round out the required set. The optional aspect_ratio defaults to 16:9. That's it - there's no seed, no negative prompt, no motion-strength slider on the Channel Edition. If you want those knobs you're looking at a different node; this one is deliberately pared down.

On queue, the node submits the request, polls the cloud, and returns three outputs: video (VIDEO - send it to a Save Video or preview node), request_id (STRING - trace/usage ID), and urls (STRING - the file location). For a beginner this is the friendliest output set in the pack: preview it, save it, move on.

Install and setup

One-time install, same as every node in this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI (or install "BizyAirPlus" through ComfyUI Manager). There are no model files - the engine is the bizyair-cloudberry and bizytrd pip packages that auto-install on startup. Then click the BizyAirPlus action-bar button to ON, get an API key from bizyair.ai, enter it when prompted (or in Settings > BizyAirPlus > API Key), and queue.

Common issues

• Nothing comes back - the switch is off, or the key is missing/invalid. The console shows the real error.
• Clips look generic - this is a text-to-video model; the prompt is doing all the work. The tooltip's advice about describing camera movement and audio cues is there because it measurably improves results.
• Billed for failures - a failed queue is still a queue attempt. There's a skip_error toggle that swaps failures for an error placeholder so the workflow doesn't abort, but the cheap fix is testing with short prompts before committing.

The pack-level caveat applies here too: BizyAir auto-installs and sends your prompts to the cloud, which makes some people nervous given ComfyUI's security history. It's a legitimate SiliconFlow product, but keep the switch off unless you're actively using cloud nodes, and assume anything you prompt is going to a server.