LTX-2: Adding outside actors and elements to the scene (not existing in the first image) IMG2VID workflow.

A developer in the AI video community has engineered a significant breakthrough for the LTX-2 model, creating a custom ComfyUI workflow that addresses a core creative constraint. The new workflow enables users to add new actors, characters, and scene elements that arrive later in a generated video sequence, even if they were completely absent from the initial reference image. This development was partly inspired by the official LTX team showcasing similar capabilities, prompting the community effort to replicate and democratize the functionality. The creator notes this solves their "biggest problem" with LTX-2 and video generation in ComfyUI without relying on specialized LoRAs, marking a step towards more controllable and dynamic AI filmmaking.

The technical approach is notably efficient, utilizing a streamlined 3-step generation process for 1080p output, which the developer found worked "just as fine" as more computationally expensive 8-step methods. The workflow is designed as an 'all-in-one' solution where users can input an initial image and specify new actors using models like Flux Klein, though the creator suggests the first parts are modular and can be swapped for other tools like NanoBanana or Qwen. This innovation represents the closest community-achieved equivalent to IPadapter's powerful reference capabilities for LTX-2, significantly expanding the narrative possibilities for AI video creators who can now craft scenes with characters entering the frame dynamically, rather than being locked to a static starting cast.