Feature Request: True Reference Image Conditioning for LTX 2.5
To the LTX development team and community,
I am writing to propose the addition of true reference image conditioning for LTX 2.5.
While I understand the model currently supports image-to-video via keyframe conditioning (using an image as the literal first or last frame), this is fundamentally different from using an image purely as a character or style reference for a new composition.
True reference image support has become a critical standard that other leading video generation models are heavily focusing on. Integrating this—whether natively would significantly streamline professional workflows. Most importantly, it would completely eliminate the need to train and manage custom LoRAs just to maintain strict character, asset, and environmental consistency across different shots and camera angles.
Is there any ongoing work to add this specific functionality via custom nodes, controlnets, or future model updates? Integrating this feature would greatly enhance LTX's utility and position it as a highly competitive tool for multi-scene production.
Thank you for your time and the ongoing development of this project.
10eros maybe,check it up