An AI company has released a set of open tools designed to let filmmakers, studios, and individual creators build their own custom AI video models trained exclusively on their own content. The update addresses one of the persistent problems in AI video generation: the inability to produce consistent results across repeated runs of the same prompt.
The company behind the tools is Lightricks, which operates under the brand LTX. Its co-founder Yaron Inger spoke exclusively to CNET about the release this week. The tools are available on GitHub as of today.
AI video generators currently produce slightly different outputs each time a user enters the same prompt. For professional creators or studios working with recognizable characters or specific visual styles, that inconsistency makes the technology difficult to use in production. LTX's framework, called LTX Trainer, is built to solve that problem by allowing users to create a model trained only on their own material.
"It was very important for us to release this, so the community can fine-tune their own data," Inger said.
Creative companies that want to experiment with AI while protecting their intellectual property can download the tools and run models locally, Inger said. Running models locally means users never have to upload their content to outside networks or cloud servers, which reduces the risk of proprietary material being exposed or absorbed by a third party's system.
The tension around intellectual property is a central issue in the creative AI space. Many artists and designers have raised concerns that AI systems are trained on their work without permission and can replicate individual styles without consent. Companies such as LTX and Adobe have responded by stating they do not train on their customers' content. Adobe introduced a similar custom model option for businesses earlier this year.
The latest update to LTX Trainer expands its capabilities beyond video. The platform can now handle audio-only training and joint audio-video training. Users can also prompt across different formats, such as generating video from audio input or generating video from a still image.
Creators can further customize their models using tools called LoRAs and IC-LoRAs. These are fine-tuned model adapters that draw on LTX's underlying AI model to replicate a user's chosen style consistently across different generated outputs. A new agentic assistant also allows users to describe what they want to create in plain language rather than through technical commands.
The release is positioned as a lower-cost alternative to the kind of AI model development that large technology companies such as Google and Meta pursue. Those companies can spend months and millions of dollars training large-scale models. LTX says its tools allow smaller creators and studios to reach similar levels of customization without those resources.
