Still getting noisy videos with New T2V Lora
Using ComfyUI native workflow with the new T2V Lora (with diff_m key removed).
CFG 1, steps 8, tried both lcm sampler and uni-pc sampler.
There are no lora key errors in ComfyUI logs, but I still get a blurry video where the rank32 T2V lora works normally.
Here the same workflow is working with the previous rank32 T2V lora.
Not working for me either. Perhaps it isn't comfy compatible?
Should be Comfy compatible, as they patched it to make it comfy compatible by removing the diff_m keys.
我也一样，不知道为什么T2V无法正常工作
please fix, the lora doesn't work.
yes, Doesnt work. Only noise output is what im getting
Hi,
Based on current testing, the complete model is functioning normally and has good dynamics, but the separated LoRA exhibits quite serious ghosting issues. We are working on resolving this issue.
Thanks.
Kijai made his own versions of the lora, I tested it and it's working fine
https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v
Kijai made his own versions of the lora, I tested it and it's working fine
https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v
have you tested all ranks and can you explain the differences? i know higher lora ranks in models leads to less compatibility does this happen here?
Hi, we have just updated the LoRA files with the correct version. Feel free to use them!
Thanks!
Hi, we have just updated the LoRA files with the correct version. Feel free to use them!
This did not solve the noisy output in my testing.