QuantStack

/

LTX-2-GGUF

like 70

Follow

QuantStack 1.71k

if you found any working workflow, please share bro

workflow?

my fault , its embedded in the video preview , open the video in comfy it will give you it

I didn't even know it was included there 🤣. But I was just making a more compact version of the workflow to share

is this only a i2v workflow?

You can also do workflows, you can search the ComfyUI examples workflows just need to use the GGUF loader

so what would cause this, i used the workflow default settings and it ran and the output looks like this?

yeah same here , results are terrible , what a waste of time

This is not something related on the quants...

Check here to see what you need to download since this PR isn't merged:
https://github.com/city96/ComfyUI-GGUF/pull/399#issuecomment-3729311217

so what would cause this, i used the workflow default settings and it ran and the output looks like this?

The cause is the ComfyUI GGUF loaders dont "read" the metadata of the GGUF models and that's something requiered by the model. Replacing the 2 files of that PR you have the support to read the metadata

Works great in comfy (been using them a lot already), but as @YarvixPA said, its not yet fully supported without pulling a PR from the gguf model loader.
Just wait a few days and it will be out in public ;-) (or pull the PR if you know how... )

And huge thanks to QuantStack for always making great GGUFs ;-)

(just a low step lower res gen for fun ;-)

is that hard to just upload a workflow?

@sniky is hard to download the example video and drop it on ComfyUI?

I get this result

Here you have a T2V, I2V and external audio workflow, full GGUF support, RAM saving with tiled vae, etc.

https://github.com/AcademiaSD/comfyui_AcademiaSD/blob/main/example_workflows/AcademiaSD_LTX-2_V24_GGUF_RES4LYF_TILED_VAE.json

i run into the error: NotImplementedError: Output channels > 65536 not supported at the MPS device.

when running the comfyui workflow attached in the sample video on a apple M1 device. does anyone know how to fix it?

I made some simplified workflows here https://huggingface.co/RuneXX/LTX-2-Workflows

I made some simplified workflows here https://huggingface.co/RuneXX/LTX-2-Workflows

thanks for sharing, i tried your workflow (I2V Basic GGUF one) and still got the error:

100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 6/6 [03:22<00:00, 33.69s/it]
100%|█████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [04:26<00:00, 88.92s/it]

Requested to load AudioVAE
NotImplementedError: Output channels > 65536 not supported at the MPS device.

i'm on pytorch 2.11.0.dev20251225

Since you tried my workflow I assume you are using Kijai's extracted split files.
For the audio vae (that errors at your end) it does need the very latest update from https://github.com/kijai/ComfyUI-KJNodes

And if you havent updated ComfyUI lately, might as well update that too

"65536 not supported at the MPS device" i never seen though, might be some apple specific error
See here : https://github.com/pytorch/pytorch/issues/144445

@RuneXX yeah i just installed comfyui-kjnodes a couple hour ago via comfyui manager, also got comfyui updated

i've also tried to export PYTORCH_ENABLE_MPS_FALLBACK=1 when starting the comfyui server but it didn't work too.

Yes, your error seems Mac related.
So what people are saying is that the fix is to update PyTorch to 2.7.x or later
https://github.com/pytorch/pytorch/issues/134416

I never installed it on mac, but its probably much the same way as you update on other platforms.

1. activate the comfy environment
2. pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org
   (or just pip install torch==2.7.1 torchvision torchaudio --index-url https://download.pytorch.org)

(i am also on 2.7.1. You could of course go to even newer PyTorch, but i usually stick to the minimum required, so that older nodes dont break if not up to date)

hooray @RuneXX , it turns out the problem is about MacOS version, i've upgraded to newer OS (Tahoe, it was Sonoma previously) and it worked.

thanks for your support and your awesome workflows too 🤟