Google Gemini Omni Flash Image To Video (Channel Edition)

Take a single image, add a prompt, and get a short clip back - Google's Gemini Omni Flash doing image-to-video, exposed as a plain ComfyUI node. This is the I2V variant of the pair: the one with an image input. Its text-to-video sibling is a separate node, but if you have a frame you like and want to see it move, this is the node.

It's part of the BizyTRD model-service family inside BizyAirPlus, SiliconFlow's cloud-execution pack for ComfyUI. Nothing here runs on your GPU. The node uploads your image, sends it to Gemini Omni Flash in the cloud, waits for the render, and hands the finished clip back as a normal VIDEO tensor you can preview or save exactly like a locally generated one.

Why you'd reach for it

Because animating a still is one of the most useful tricks in ComfyUI, and doing it locally usually means wrestling with Wan or LTX, downloading multi-gigabyte checkpoints, and tuning schedulers until your VRAM gives up. This node swaps all of that for an API call. If you've already generated a strong frame locally and want to bring it to life - a character turn, a camera push-in, water moving behind a landscape - this gets you there without touching a single model file.

The practical trade you're making: money and latency for convenience. It's a closed model, billed per generation, and the community's stance on closed video models is well documented - Veo, Sora and friends sit ahead of open alternatives but are exactly the thing open-source efforts like Wan keep trying to close the gap on. When you need a result now and don't own the hardware, this is a legitimately good stopgap.

How it works

Wire an image into the images input (a COMFY_AUTOGROW_V3 port, so it accepts a connected image gracefully), write your prompt, pick resolution (720p by default) and duration (default 4), and queue. The prompt field accepts 1–2048 characters, and the tooltip tells you exactly how to spend them: describe the scene, characters, action, environment, camera movement, screen style, and audio cues in detail. That last one is not filler - Gemini Omni is an omni model and detailed prompts genuinely move the output.

Optional aspect_ratio defaults to 16:9, but the tooltip notes the system auto-adapts the ratio to your first frame when you upload one, so you can usually leave it alone.

Three outputs: video (VIDEO - wire this to a Save Video or preview node), request_id (STRING - trace ID), and urls (STRING - where the finished clip lives, if you want the file path or to host it elsewhere).

Install and setup

The node ships with BizyAirPlus, so it's the same one-time install 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" via ComfyUI Manager). No model downloads - the pack's real engine arrives as pip packages (bizyair-cloudberry, bizytrd) auto-installed on startup. Then click the BizyAirPlus button in the action bar so it's ON, grab an API key from bizyair.ai, and queue.

Common issues

• Video never appears - check the switch is ON and the key is valid. If a call fails, the ComfyUI console has the real error.
• Stuck on a bad first frame - the model starts from what you give it; a blurry or oddly-cropped input makes a blurry clip. Crop and clean the source before wiring it in.
• Cost surprises - every queue is a billed generation. The skip_error toggle (set it to output a placeholder instead of aborting on failure) is the only failure-handling input, so treat failures as a bill you want to avoid, not just an error you fix.

And the standing caveat for this whole pack: BizyAir auto-installs and sends your inputs to the cloud, which makes some people uneasy - fairly, given ComfyUI's history with sketchy custom nodes. It's a legitimate SiliconFlow product, but flip the switch OFF when you're only running local workflows, and be deliberate about what you feed it.