noob question
im new here, should i use this model or the Unsloth GGUF model ? what are the different ?
It would depend on your hardware. You should use smaller GGUFs only when you absolutely must, as in when you simply don't have enough VRAM/RAM to run it. Otherwise fp8_scaled will currently work better with ComfyUI memory management as it's natively supported, while GGUF support is through custom nodes, and especially if have Nvidia GPU that supports fp8 on hardware level (40xx and up) fp8 will be faster.
There's also an argument that GGUF Q8 can be slightly better quality, but that's something to judge for yourself.
LORD KIJAI, DO YOU HAVE ANY WORKFLOW FOR THIS FITTING FOR A 4090?
You can use any LTX workflow, but swap out the main model loaders to fit the use of the split models here.
See image on front page https://huggingface.co/Kijai/LTX2.3_comfy
And should you be stuck, you can always try one of mine here
For 4090 you can try the fp8_input_scaled version, which said to be more than 50% faster than fp8_scaled (according to someone test at reddit).
50% faster
I guess i soon gotta get a new gpu ;-) sounds like it might be quite a performance boost