LTX 2.5 Alpha Gen: ComfyUI setup and compositing

Jonathan Winbush demonstrates LTX 2.5 Alpha Gen in ComfyUI to generate alpha mattes from existing footage on his laptop. The tutorial covers a local AI workflow for extracting subjects, including soft hair edges and partial transparency. Alpha Gen is in beta.

Setup starts with LTX's JSON template and downloads for the text encoders, LoRAs, diffusion models and VAEs, placed in their respective ComfyUI model folders. Winbush loads footage into the video node, adds a blank image to the image node and disables image input before running the workflow. He then imports the matte and original footage into After Effects for compositing.

Examples include glass, smoke, backlit curtains and uneven green screens. Winbush judges the results favorably against his earlier keying and manual rotoscoping work; these are his assessments of the demonstrated clips.

Resolution and duration need attention. He explains that image dimensions use multiples of 32 and that output length follows an eight-frames-plus-one constraint. One matte ends before its source clip. A square example produces 544 by 544 output. For higher resolution, he changes the source-video subgraph's shorter-side setting to 1080. That roughly six-second example takes about 11 minutes on his laptop. The transcript gives no laptop specifications, so those timings are not a hardware benchmark.