✦ LTX-2.3 LoRA Trainer

Train a LoRA / IC-LoRA on your own videos — runs on HF Jobs, pushed to your Hub.

Training mode

IC-LoRA (in-context control) learns from pairs — a target clip clip.mp4 and its control video clip_reference.mp4 (depth, pose, edges, an inpainting-masked version, …), matched by filename. T2V / I2V use only the target clips.

Dataset source
ℹ️ Captioning tips
  • Drop a clip.txt next to each target clip for a per-clip caption. The “same caption for all” box is used only where a clip has no .txt.
  • LTX-2.3 likes long, detailed, chronological captions (~200 words): describe motion, camera, lighting, and any audio.
  • The trigger word is prepended to every caption so you can invoke the LoRA at inference.
  • For IC-LoRA, also upload each clip_reference.* (matched by filename).