verifier.org

LTX-2.3

the best open-weight video model, with a licence that isn't what you think

free to self-host; $0.06 / second at 1080p on fal#self-hosted-pipelines-ne#4k#audio-in-one-pass

twenty-second clips at up to 4K with genuine single-pass audio, running on your own hardware — just don't believe anyone who tells you it's apache 2.0.

ltx-2.3 is a 22-billion-parameter audio-video model from lightricks, and it is comfortably the strongest open-weight option here: leader among open weights on artificial analysis' image-to-video with audio board, twenty-second clips — the longest of any mainstream model — up to 4K, and 24 or 48fps.

it generates synchronised video and audio in a single forward pass, which most open models still can't do at all.

the description above is ours, condensed from the ranking. pricing moves — check it on the vendor's own page before you rely on it.

pricing
free to self-host; $0.06 / second at 1080p on fal
website
github.com
our verdict

twenty-second clips at up to 4K with genuine single-pass audio, running on your own hardware — just don't believe anyone who tells you it's apache 2.0.

we researched this category against vendors' own pricing pages and licence files. that is where this line comes from — not from the vendor, and not from anything they paid for.

more ai video models

LTX Studio

storyboard-to-video workspace from Lightricks, built around its own LTX models

#storyboard#workspace

Moonvalley

video model trained on licensed footage, aimed at film production

#licensed-data#film

Vidu

text and reference-to-video model from Shengshu, with a consistency feature for characters

#reference-to-video

Gemini Omni Flash

we checked this$0.10 / second of 720p, direct from google

first on every leaderboard, and one of the cheapest

#almost-everything#until-you-hit-the-ten-se