the new brand from ltx video v0.9.5
Hello, there is a brand new model. Can you create a model on gguf with the ltx video v0.9.5? That would be very nice if you do that, thanks.
https://huggingface.co/Lightricks/LTX-Video/blob/main/ltx-video-2b-v0.9.5.safetensors
calcuis already made the gguf: calcuis ltx 0.9.5 gguf, but since it uses a new architecture (pig architecture, reference) and city96 didn't update his code yet (to use the new architecture), you need to use calcuis' node: calcuis node.
Edit:
You still need the vae (it isn't included in the gguf), but to save space you can extract it from the original model with the comfyui default vae save node:
then you can select the newly extracted vae (from the output folder) and convert it to fp8 with calcuis' tensor cutter node:
Overview:
Original model (unet + vae): 6 gb (rounded)
gguf unet Q8: 2 gb (rounded)
vae fp8 (extracted & cutted): 1,2 gb (rounded)