yuvraj108c/LTX-2.3-22b-IC-LoRA-Any-Trajectory-Instruction

LTX-2.3 22B IC-LoRA Any Trajectory Instruction (ATI)

This is an experimental IC-Lora for porting Any Trajectory Instruction (ATI) to LTX using minimal resources.

Model Details

• Base Model: LTX-2.3-22b Video
• Training Type: IC LoRA
• Control Type: Video with splines in locations where motion tracking is given as a constraint
• Bucket Size: 768x768x81
• Training Steps: 3000
• Learning Rate: 1e-4
• Dataset Size: 25 video pairs
• Hardware: 1 H100
• Training Time: < 5 hours

Using in ComfyUI

1. Download ltx-2.3-22b-ic-lora-any-trajectory-instruction-3000.safetensors into models/loras.
2. Use the provided IC-LoRA workflow
3. Draw motion paths on input image, ensure you use the following settings:

Limitations

• First experimental IC-LoRA checkpoint trained on only 25 video pairs
• Motion points occasionally appear on the results at different resolutions/frame count

Acknowledgments

• LTX-Trainer: https://github.com/Lightricks/LTX-2/tree/main/packages/ltx-trainer
• ATI: https://github.com/bytedance/ATI
• FL Path Animator Node: https://github.com/filliptm/ComfyUI_Fill-Nodes

Support

Consider supporting me if you want continuous updates and new IC-Loras!