***UPDATE*** 26/6/2026 I just uploaded GGUF and Full model workflows for low VRAM.
LTX Director Motion Brush V2 is a ComfyUI workflow/node pack for LTX 2.3 timeline generation with per-image Motion Brush control.
Draw motion tracks directly on storyboard images, convert them into sparse IC-LoRA motion-control guide video, and build LTX 2.3 shots with Director-style timing, retake workflows, matte clips, per-image guide strength, and motion-carry transitions.
Nightly preview release: best for users comfortable testing new ComfyUI workflows.
Requires:
ComfyUI-LTXVideo
comfyui-kjnodes
ComfyUI-Impact-Pack
LTX 2.3 model setup
LTX 2.3 IC-LoRA motion-track-control + IC-LoRA-ingredients
Repo:
https://github.com/exportAnything/ComfyUI-LTX-Director-Motion-Brush
Description
FAQ
Comments (3)
torch.OutOfMemoryError: Allocation on device This error means you ran out of memory on your GPU.
That actually happened to me once with this worflow + node, and assumed it was a one-off because it worked fine afterwards. Ill look into this for sure.
I found a very plausible cause, and it is mostly the shipped workflow defaults, not the node pack install itself.
The current example workflow is heavy for a first run:
It defaults to 960x544 at 360 duration frames, about 15 seconds;
It runs Stage #1 and then Stage #2 upscale by default;
Stage #2 loads the ltx-2.3-spatial-upscaler-x1.5 model.
Lastly, there are three active SaveVideo outputs, so the workflow is not just doing one simple generation.
ComfyUI’s OOM handler unloads loaded models after an OOM, so a reset clears cached models/fragmentation and may make the next attempt squeak through with lower VRAM.
You can replace the default workflow settings with 120 frames instead of 360 and:
lower first-run resolution, maybe 768x448 or 512x288 ;
And/or disable Stage #2 upscale by default;
And/or disable stitched comparison output by default;
And/or keep only one active SaveVideo output.
If you'd rather wait, I'm preparing a "first-run safe" workflow that I'll be uploading sometime tonight.