plz force update kjnode, ltxv(comfyui), gguf
v1.3
ltx2 new samepler node update,
The overall sound has been significantly improved. Image quality doesn't seem to have changed much.
Also, the new sampler requires more image strength in i2v.
plz force update kjnode, ltxv(comfyui), gguf
v1.2
some fix.
The gguf model used has changed. If possible, please use "ltx-2-19b-dev-Q8_0.gguf" by unsloth
v1.1
We've heard that Kijai's video_vae has been changed due to an som issues. The vae loader has also changed. The link remains the same, so please download it again.
And you also need to force update kjnode.
This is the workflow I personally use.
All necessary information is noted within the workflow.
Both T2V and I2V are possible.
It runs with similar specs to Wan2.2 in the same computer environment, same video resolution and video length.
In my computer environment,
Wan2.2-I2V-A14B-Q8_0 8step 16fps 300sec
ltx-2-19b-Distilled_Q8_0 8step euler 24fps 150sec
ltx-2-19b-Dev_Q8_0 20step euler 24fps 300sec
ltx-2-19b-Dev_Q8_0 20step res_2s 24fps 450sec
I have 24gb vram, 32gb system ram, but it runs fine with -novram, so it should work fine with 8gb vram.
If you run it with -novram, the generation time will be about twice as long.
If possible, keep your prompts as long as possible. If they're too short, errors may occur or unintended effects may be added. Be sure to include sufficient scene descriptions and dialogue.
Description
v1.2
some fix.
The gguf model used has changed. If possible, please use "ltx-2-19b-dev-Q8_0.gguf" by unsloth
FAQ
Comments (5)
I'm having a hard time using this. Everything I try results in a noisy mess. This is the case for both T2V and I2V, across both 8-step and 20-step mode. My models are the exact same as listed in the workflow note.
In such cases, it's usually due to the KJNode and GGUF custom node versions. Both appear to be up-to-date, but you'll need to force an update.
That worked, thanks! Also, I just want to mention that the LTX team released an update to their custom nodes. They added a new "LTXVNormalizingSampler" which is a drop-in replacement to the existing sampler. It seems a lot better, so it might be worth looking into
@Unhing3d Thank you. I will try it.
@Unhing3d While audio and image breakup have definitely improved, image quality has deteriorated. I need to explore ways to utilize the new node.