Ltx 2.3 Text To Video

Here's the honest pitch for this node: LTX 2.3 is a 22-billion-parameter video model whose official minimum is a 32GB+ GPU, and the community's realistic advice is "quantize it and pray about your system RAM." It is genuinely one of the best all-round locally runnable video models out there - but "locally runnable" is doing a lot of heavy lifting. This BizyAirPlus node is LTX 2.3 without that sentence: the open-weight model runs on BizyAir's cloud, you keep the ComfyUI workflow.

What it is

This is one of the pack's "Self-Hosted" category nodes - meaning BizyAir runs the model itself rather than channeling a vendor API. LTX 2.3 is Lightricks' open-weight video model (March 2026), famous for native synchronized audio, strong prompt adherence, and being the best speed-to-quality balance in open video. This node hands you its text-to-video mode over HTTP: prompt in, finished video out, zero VRAM involved.

How it works

You get the same LTX 2.3 pipeline the local nodes use - text encoder, sampling, the audio-sync'd decode - but executed remotely. What you don't get is any of the fiddly local setup: no 22B checkpoint, no Gemma-class text encoder eating RAM, no VAE troubleshooting. The tradeoff is the same as every BizyAir node: it's metered, and you don't get sampler-level control.

The inputs that matter

• prompt - the creative input. LTX is prompt-sensitive; describe the scene, the action over time, and the sounds you want. Short, specific lines beat essays.
• resolution - defaults to 1080P, which is the sensible setting.
• duration - locked to 5 seconds on this node. LTX is a speed tier, not a long-form tier, and 5s is its happy place.
• display - presentation mode, default "horizontal."
• seed - set for reproducibility, -1 for random.

api_config and skip_error are the standard pack options.

Outputs

video wires into SaveVideo or a preview. request_id is the cloud job ID. urls gives direct links to the rendered files.

Installing

ComfyUI Manager → search "BizyAirPlus", or:

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

Restart, toggle the BizyAirPlus button ON, enter your BizyAir API key from bizyair.ai. No model downloads - that's the point.

Where people get burned

Expectation management. If you're comparing this against a local LTX workflow you've tuned, you're giving up the fine controls and paying per clip - this node is for the other 90% of the time: when you want LTX's output without babysitting 80GB of VRAM, or when you're on hardware that can't run 2.3 at all. Also, it's a 5-second model here, so don't plan a 30-second master shot on it - that's not what this tier is for. And LTX rewards good prompts; a lazy prompt gets you a technically clean video of nothing in particular.