Overview
This is a straightforward workflow for the LTXV 13b Distilled model, designed for quick and easy video generation. The distilled model runs much faster than the full version, but still delivers impressive quality-making it a great choice for rapid prototyping or when you want to generate a lot of videos quickly.
How It Works
Scheduler Steps:
You can set the scheduler to use anywhere from 4 to 12 steps.For image-to-video (i2v) tasks, using 12 steps usually helps retain more detail from your input image.
8 steps is a good general-purpose setting-fast and still pretty accurate.
If you go lower (e.g., 4-6 steps), the workflow is even faster, but with complex images, the output might drift visually or change style.
LoRA Support:
There’s a Load LoRA node included. Enable it if you want to add a LoRA for extra effects or style-otherwise, just leave it off for standard generation.Speed:
On an H100 GPU, this workflow generates a 5-second video (8 steps) in about 11 seconds.Combining Models:
For the best results, you can combine the distilled model with the full LTXV 13b model-use the distilled version for speed, and the full model for final, high-quality renders if needed. But honestly, the distilled model alone is already pretty solid for most use cases.