Ltx 2.3 Image To Video

Image-to-video is where LTX 2.3 went from "promising" to "the one people actually finish work with." The 2.2-era complaint - that I2V was a static nightmare - got fixed, hard. This BizyAirPlus node is LTX 2.3's I2V mode served over the cloud, so you can animate a still with one of the best open video models of 2026 without owning the GPU it officially wants. For anyone who's already got a strong image (from Z-Image, Klein, whatever) and needs it to move, this is the shortest path in the pack.

What it is

The "Self-Hosted" LTX node - BizyAir runs Lightricks' 22B open-weight model, you consume it as an API call. Text-to-video's sibling, and arguably the more useful one: feed it an image and a prompt and get back an animated clip that keeps the subject, with LTX 2.3's native audio on top.

How it works

The image input is your starting frame, the prompt describes what happens next, and the cloud runs the LTX 2.3 I2V pipeline - conditioning on your image, sampling the motion, decoding audio-synced video. Same model the local nodes run, none of the setup. You control resolution, duration, orientation, and seed; the sampling itself stays remote.

The inputs that matter

• image - required. The still to animate. Garbage in, garbage moving - the source image quality is the ceiling.
• prompt - describe the motion over time, including any sound. LTX rewards specific, physical descriptions.
• resolution - 1080P by default.
• duration - 5–10 seconds (note the I2V node gives you up to 10, unlike the T2V node's fixed 5).
• display - presentation mode, default "horizontal."
• seed - reproducibility.

api_config and skip_error are the pack standard.

Outputs

video → SaveVideo/preview. request_id is the cloud job ID. urls are direct links to the rendered clips.

Installing

ComfyUI Manager → "BizyAirPlus", or:

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

Restart, turn the BizyAirPlus button ON, enter your API key from bizyair.ai.

Where people get burned

The source image is 80% of the result - LTX I2V respects the still, and a muddy, low-contrast frame animates into muddy motion. Feed it a clean, well-exposed image. The duration range is 5–10s; pushing 10 with a lot of motion invites the smudgy movement LTX critics complain about, so keep actions modest on long takes. And remember it's metered: iterate on prompt at 5s before spending a 10-second render, and check that you're not double-paying by animating an image you'll reroll anyway.