Hippotes

/

LTX-2.3-quants

like 32

fp8mixed and nvfp4mixed are up :)
idk if I will do non-mixed ones, the quality hit seems not worth saving a few gigs or 1-2s.

fp8mixed and nvfp4mixed are up :)
idk if I will do non-mixed ones, the quality hit seems not worth saving a few gigs or 1-2s.

genius, many many thanks

fp8mixed and nvfp4mixed are up :)
idk if I will do non-mixed ones, the quality hit seems not worth saving a few gigs or 1-2s.

i know you said it might not be worth, but your version of it let's me have more speed in wan2gp with some weird options, i ask here you to reconsider a bit and maybe do the full nvfp4 of the 1.1 as you did with the original
if you don't want to it's okay, but never hurts to ask, thanks for the others they work good

Hi! Where can I download the new workflow for
ltx-2.3-22b-distilled-1.1-nvfp4mixed.safetensors

fp8mixed and nvfp4mixed are up :)
idk if I will do non-mixed ones, the quality hit seems not worth saving a few gigs or 1-2s.

i know you said it might not be worth, but your version of it let's me have more speed in wan2gp with some weird options, i ask here you to reconsider a bit and maybe do the full nvfp4 of the 1.1 as you did with the original
if you don't want to it's okay, but never hurts to ask, thanks for the others they work good

welp, I can't vouch for the quality but weights are up: 1.1-nvfp4 and 1.1-fp8

Hi! Where can I download the new workflow for
ltx-2.3-22b-distilled-1.1-nvfp4mixed.safetensors

1.1 is only a better distillation of dev model, drop-in replacement in your existing workflows

fp8mixed and nvfp4mixed are up :)
idk if I will do non-mixed ones, the quality hit seems not worth saving a few gigs or 1-2s.

i know you said it might not be worth, but your version of it let's me have more speed in wan2gp with some weird options, i ask here you to reconsider a bit and maybe do the full nvfp4 of the 1.1 as you did with the original
if you don't want to it's okay, but never hurts to ask, thanks for the others they work good

welp, I can't vouch for the quality but weights are up: 1.1-nvfp4 and 1.1-fp8

Hi! Where can I download the new workflow for
ltx-2.3-22b-distilled-1.1-nvfp4mixed.safetensors

1.1 is only a better distillation of dev model, drop-in replacement in your existing workflows

thanks brov, seriously, thanks

1.1 is only a better distillation of dev model, drop-in replacement in your existing workflows

Did you also make new Distilled Lora 1.1 with lower rank and size?

1.1 is only a better distillation of dev model, drop-in replacement in your existing workflows

Did you also make new Distilled Lora 1.1 with lower rank and size?

Kijai did: https://huggingface.co/Kijai/LTX2.3_comfy/tree/main/loras, using adaptive rank, best it can be

1.1 is only a better distillation of dev model, drop-in replacement in your existing workflows

Did you also make new Distilled Lora 1.1 with lower rank and size?

Kijai did: https://huggingface.co/Kijai/LTX2.3_comfy/tree/main/loras, using adaptive rank, best it can be

Thanks.
By any chance you know with which strength I need to use this lora? Is it 1.0, 0.5 or other?

Thanks.
By any chance you know with which strength I need to use this lora? Is it 1.0, 0.5 or other?

I'm not sure... It is meant to be used at 1.0, that should bring dev + lora closest to the distilled model. But even official ComfyUI template uses 0.5, I have seen gens with 0.6~0.7 that are less saturated and with more natural motion, and some 2 stages workflows use 0.4 -> 0.7
I often stick with the distilled model for that exact reason of not having to guess or trial & error the correct strength for the specific thing I'm generating at the time