Lightricks has released IC-LoRA (Image/Video Conditioning LoRA) for the LTX-2.3-22B model, which allows for the automated creation of "clean plates" by removing people, vehicles, and other moving objects from video sequences while preserving background details.
What Happened
The developed IC-LoRA module enables video-to-video reconstruction by using a reference video as a control signal. This eliminates the need for text trigger words to manage the process. The tool allows for cleaning videos of dynamic objects while maintaining the architectural integrity and details of the static environment.
Context
The technology is based on the Image/Video Conditioning (IC) method, which provides more precise generation control through visual references. This represents an alternative to the traditional process of creating complex masks and manual rotoscoping in VFX pipelines.
Why It Matters for the Industry
For the visual effects and compositing industry, this signifies a shift from labor-intensive, pixel-by-pixel manual work to semantic layer management. Automating clean plate creation simplifies workflows and allows such methods to be integrated into cloud APIs and commercial video editors, creating new standards for "smart" compositing.
Why It Matters for Users
Regular users and content creators gain an accessible way to professionally "clean" video footage of accidental passersby or cars. To do this, one only needs to describe the desired empty landscape in a prompt, without requiring deep knowledge of masking or complex compositing software.
What Is Not Yet Known / Limitations
There are uncertainties regarding the computational cost of inference and latency when working with a 22B parameter model, which may affect the feasibility of mass real-time implementation.
Sources
Author
Look at AI, Editorial Team