LTX 2.3 image-to-video space.
runs an in-process comfy workflow against the LTX 2.3 fp8 checkpoint. the underlying model is LTX 2.3 audio-video, so audio is generated alongside the video and muxed into the output by default. an audio toggle is exposed if you want a silent mp4.
quick usage:
- upload a starting image
- write a prompt
- pick a preset (
tunedis the default) - generate
face targeting modes:
anchor only: broad latent anchor, no per-face guideauto face: mediapipe face detection, likeness guide + anchormanual bbox: pastex1,y1,x2,y2normalized 0-1
keep generations short on first try. larger frames and longer durations spend more zerogpu time.