YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

LTX-Video 2.3 β€” DMD + Distill v1.0 Hybrid LoRAs

A set of experimental hybrid LoRAs for LTX-Video 2.3 (22B).

❓ Why these exist

Testing different LTX 2.3 LoRAs showed a clear problem:

  • DMD LoRA & Distill v1.1: Great fast movement, but burns colors and contrast.
  • Distill v1.0: Great picture quality and natural colors, but slow/sluggish motion.

To combine their strengths, I took DMD for the movement (first blocks) and replaced the end blocks with Distill v1.0 at 0.6 strength to fix the colors.


πŸ”— Credits & Source Models


πŸ“¦ Model Versions

The name format is: ltx23_dmd+d10_b[BLOCKS]_s06.safetensors

  • b45-46: Maximum DMD movement. Replaces blocks 45–46.
  • b45-47: Maximum DMD movement, with a fix for color burn on final output blocks.
  • b35-47: Balanced mix between DMD movement and Distill v1.0 colors.
  • b25-47: Smoother movement, cleanest colors (shifted further towards Distill v1.0).

(Note: These descriptions are purely theoretical based on block weights. Real-world testing is required).


βš™οΈ How to use

  • Steps: 4 to 8 steps
  • CFG: 1.0
  • LoRA Strength: 1.0 (or slightly lower, e.g. 0.8–0.9, depending on the specific merge)
  • Audio: Audio layers are kept from DMD.

These models are experimental and need testing. Try them out and share your feedback!

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support