LTX 2.3 Image to Video
Image to Video

LTX 2.3 is an iteration of the open-source LTX-Video model family from Lightricks, discussed mainly for its layered merge image-to-video workflow — generating motion in passes and combining them so the subject from your source image stays consistent across the clip. Like the rest of the LTX line it ships as open weights you can run locally with tooling such as ComfyUI or diffusers, rather than as a hosted consumer product.

No. VO3 does not host LTX 2.3, and we would rather say that plainly than sell you a page that implies otherwise. VO3 runs a browser-based image-to-video pipeline on Veo 3.1, Sora 2, Kling 3.0, Wan 2.7, Hailuo and Seedance 2 — the same job of turning a still frame into a moving shot, without a local install.

The weights are open, so downloading and running LTX 2.3 costs nothing in licence fees. The real cost is hardware and time: a GPU with enough VRAM, a working Python and CUDA environment, model and LoRA downloads, and the hours spent maintaining it. Check the current LTX licence on the official repository before any commercial deployment, since open weights and unrestricted commercial rights are not the same thing.

Requirements shift with each release and with the resolution and clip length you target, so the model card on the official repository is the only figure worth trusting. As a general rule for open-source image-to-video: a modern NVIDIA card with plenty of VRAM, and the willingness to trade resolution or frame count when you run short. If that sentence sounds like a project rather than a Tuesday, hosted generation is the shortcut.

It depends on what you optimise for. LTX 2.3 wins on control, privacy, custom fine-tunes and marginal cost once the hardware is paid for. Hosted models win on setup time, native audio, high resolution without VRAM limits, and never debugging a dependency at midnight. Studios frequently run both — open-source for bespoke pipelines, hosted image-to-video for everything on a deadline.

Open the VO3 create page, upload your image, write one line describing the motion you want, pick a model and resolution, and render. The whole loop runs in the browser on any laptop — no GPU, no CUDA, no ComfyUI graph. New accounts get free credits, so you can compare the output against your local image-to-video results before spending anything.

Yes. Videos generated on VO3 paid plans can be used in ads, product pages, social posts and client deliverables. If you are running LTX 2.3 yourself instead, read that model's own licence terms — open weights do not automatically grant unrestricted commercial use, and the terms differ between open-source video models.

Standard JPG and PNG uploads work. Feed the model the highest-quality version of your frame you have, with the subject clearly lit and unobstructed, and choose the aspect ratio you will actually publish in — 9:16 for TikTok and Reels, 16:9 for YouTube and site headers, 1:1 for feed placements. Cropping a widescreen render into vertical afterwards loses more than starting vertical does.