LTX2.3 Control & Abliteration! (Patreon)

Video - https://youtu.be/Y-ZLRAhPyLk

Overview

As an alternative to the individual ControlNet LoRA files, Lightricks have now also released a 2.3 version of their Union Control LoRA (supporting things like pose, canny & depth).

https://huggingface.co/Lightricks/LTX-2.3-22b-IC-LoRA-Union-Control

This LoRA needs a slightly different loader (as it does some scaling), so I've added this option into my LTX 2 workflow for you :)

The link to the Union LoRA file has been included in the model download note:

You might also have found that the LTX2 Prompt enhancer sometimes does things like this:

... which is where the abliterated LoRA comes in! There are a couple of them available, with links provided directly in the workflow:

You may also be interested in Kijai's "v2" scaled model - https://huggingface.co/Kijai/LTX2.3_comfy

Results between the model are all fairly similar, so test away and see which you prefer.

Workflow Information

You will need to install a number of custom nodes, as detailed in the workflow note:

If, for whatever reason, you're having trouble installing the Radiance nodes, they can simply be replaced with your favourite alternative for image upscaling and image effects (such as the native sharpen node, KJ Node's colour match, etc). See the -NoRad workflow as an example.

When opting to use the Union LoRA, you'll need to:

Enable the Union LoRA Loader (disable when not in use)

Enable Pose or Canny & provide your input video

Note that rescaling now happens AFTER video load:

I haven't tried loading 4k videos directly, but if it does cause OOM issues on your hardware, then you can still set the custom_width & custom_height directly.

For the Abliterated CLIP LoRA, simply enable if the LTX2 AI Prompt Enhancer says "no":

I've also added an IMG Vision switch too, making it much easier to ignore images if using either of the prompt enhancers:

If you do want to use the image, simply keep it enabled:

As an alternative to using IMG1 or IMG2 for the AI Prompt Enhancers, you could also use the first pose video frame:

You can update the GET node with your desired image option:

Additionally, I've also added a Stage 2 Sampler option in the main setting panel:

euler_ancestral_cfg_pp is a decent default, but play around and see what works best for you!

Enjoy! :)