Has ltx-2-3 been released?
https://ltx.io/model/ltx-2-3
This webpage has already been published, we're just waiting for Huggingface to upload the model
they say there would be an update real soon, but when I dont know ;-)
Looks ComfyUI is added support today, so I guess it coming.. . ;-)
- New VAE with sharper details
- Cleaner sound
- fixed 9:16 (portrait) mode
- new camera loras
- better prompt understanding
- last frame interpolation
As far as i can tell ;-)
But the model file size of 46GB is too large, I'll just wait for FP8
But the model file size of 46GB is too large, I'll just wait for FP8
That will come for sure ;-) the community will make quantized models, ComfyUI and Kijai as well i am sure ;-)
I don't know if the UNet loading method has been optimized, but the effect of our experimental video generation is not as good as the previous node model
I don't know if the UNet loading method has been optimized, but the effect of our experimental video generation is not as good as the previous node model
Seems pretty much same here, perhaps GGUF is a bit slower (but i didnt do a comparison)
I haven't upgraded to Comfyui or kjnodes, I don't know if I can run it
If you could run it in the past, the upgrade isnt dratical. Its basically just using different node for loading the main model.
The upgrade is very minor, and should be fine ;-)
Kijai works faster than speedo gonzales, the split models are already up:
https://huggingface.co/Kijai/LTX2.3_comfy
Will make some easy workflows for the new models (not a huge difference from before, but a bit simpler model loading )