LTX-2.3 GGUF with GGUFLoaderKJ

[RuneGjerde]

You might also upload something for GGUF later on, but gave it a try already with the Unsloth variant.
At first glance it looked somewhat similar to your logic, with GGUF + embeddings. But getting errors.

A lot of size mismatch for transformer_blocks.

I know its not your responsibility but thought i'd mention it just in case you wanted to be compatible with the Unsloth upload.
https://huggingface.co/unsloth/LTX-2.3-GGUF

And if you are already preoccupied with other things no rush, and if you want to skip it, understandable, since its not your models ;-)

(but could also be i did something wrong... although i looked over several times, but maybe i need a coffee.. :-p )

UPDATE:
They work now, after they been re-uploaded. ;-)

[RuneGjerde]

ah .. i see someone else also tried #560 ;)

[kijai]

Does it work with https://github.com/city96/ComfyUI-GGUF ?

[RuneGjerde]

Does it work with https://github.com/city96/ComfyUI-GGUF ?

No. Might be where the errors stem from since i guess you node is ultimately based on that
Will try some other ggufs too, maybe its Unsloth ones that are not compatible with City's gguf loader.

I saw VantageAI say his loads fine, so will try those.
(just for the super low vram users, the fp8 ones run fine my end)

[Djanghost]

Does it work with https://github.com/city96/ComfyUI-GGUF ?

No, I tested all the GGUFs Q8 currently available on huggingface (even the one from
QuantStack before it was removed). I got the same error on all of them, with your node and the UNET GGUF loader (city96).

[RuneGjerde]

VantageAI's GGUF loaded just fine. So i guess the QuantStack (now re-uploaded, might be working) and Unsloth ones are a bit different.

[kijai]

Does it work with https://github.com/city96/ComfyUI-GGUF ?

No. Might be where the errors stem from since i guess you node is ultimately based on that Will try some other ggufs too, maybe its Unsloth ones that are not compatible with City's gguf loader.

I saw VantageAI say his loads fine, so will try those. (just for the super low vram users, the fp8 ones run fine my end)

Yeah I'm quite literally just using that with some extra options in my node.

[RuneGjerde]

Yeah I'm quite literally just using that with some extra options in my node.

Yes. Maybe City69 will update ;-) if needed.

[RuneGjerde]

No, I tested all the GGUFs Q8 currently available on huggingface (even the one from
QuantStack before it was removed).

The new re-uploaded QuantStack ones works fine now
(and links back to Kijai's upload for text encoder etc. So i guess the intention was to work with comfy from the start)
https://huggingface.co/QuantStack/LTX-2.3-GGUF

[RuneGjerde]

Thought I'd just add that the Unsloth GGUF models now also works great, after a re-upload ;-)
https://huggingface.co/unsloth/LTX-2.3-GGUF

(since they were mentioned earlier as not working, they are all fine now)

[Lim-Kid]

Thought I'd just add that the Unsloth GGUF models now also works great, after a re-upload ;-) https://huggingface.co/unsloth/LTX-2.3-GGUF

(since they were mentioned earlier as not working, they are all fine now)

Well it worked with city96's one but not with KJ's (the latest version from March 31/2026)