You need to put NAG (negative attention guidance) between the LightX2V Lora node and Scheduled CFG Guidance nodes. The negative prompt is currently placebo and does nothing with the default CFG 1.0 settings.

drsammydJan 9, 2026

Is this for the I2V or FLF2V workflow?

aureliusJan 9, 2026· 1 reaction

@drsammyd As long as your CFG is 1, as it is with lightx2v, you need NAG for the negative prompt to do anything. Doesn't matter what kind of generation it is. You need two NAG nodes, one for high and one for low. If using SVI, you need them for every video in the chain.

mifink94Jan 4, 2026

CivitAI

hello)

is version of Comfy important? I cant install itWanImageToVideoSVIPro, UnetLoaderGGUF :(

Legendaer

Author

Jan 4, 2026

Hi, you need the nightly version of ComfyUI-KJNodes for the WanImageToVideoSVIPro node since its still very new. For the gguf unet loader you need ComfyUI-GGUF

mifink94Jan 4, 2026

@Legendaer thanks boss, I think I understand. should I either wait a little longer or install the nightly version of KJ?

Sorry for the stupid questions. For some reason, learning Comfi is really hard for me...really hard(

Legendaer

Author

Jan 4, 2026

@mifink94 Well, that depends on how patient you are, since it’s uncertain when the version number will be incremented.

mifink94Jan 5, 2026

@Legendaer Its clear, thank you very much for help!)

SiMBa25Jan 5, 2026

CivitAI

Hello, I'd like to ask why SageAttetion isn't working. Everything works fine in other similar workflows.

I have SageAttetion 2.2.0 and Triton 3.5 installed.

And one more question: how can I speed up the animation? It's too slow.

aureliusJan 6, 2026

Try to lower Shift and make the prompt more dense or reduce the video length per segment. As for SageAttention, no idea why it wouldn't work if it works elsewhere.

Legendaer

Author

Jan 10, 2026

Also not sure why sageattention wouldn't work for you if it works elsewhere since I haven't had any issues myself. And to speed up the animation, you can try what the other comment suggested. But if it's just for a flat out speed increase just increase the fps. Also, the lightx2v lora tends to reduce movement, so you could either disable it, which would require a lot more steps, or use the 3Sampler workflow, where the first step is without the lora.

SiMBa25Jan 11, 2026· 2 reactions

@Legendaer I apologize for my inattention...

Everything works, it's just that in automatic mode, I selected options that I don't have in Python.
I only have sage attention 2.2.0 and triton-windows 3.5.1 installed.
I just needed to manually select sageattngk int8pv_fpl6_triton since I don't have the other options...
Now everything works great.
I use DasiwaWAN22I2V14BTastysinV8_q6High.gguf and DasiwaWAN22I2V14BTastysinV8_q6Low.gguf

SiMBa25Jan 11, 2026· 1 reaction

@Legendaer And yes, I use the 3 KMsampler, I liked it better too.

organic10255121Jan 16, 2026

CivitAI

Thank you for the great workflow!

But is setting "start percentage=0" in every sampler fine?

Isn't it the same as setting "start at=0" in general high and low K samplers?

NemesisEnforcerFeb 20, 2026· 3 reactions

CivitAI

Excellent stuff, thank you! My next mission in life, acquire more RAM and VRAM. Then I will be complete.

bevangg956Apr 12, 2026

There are some things in life that you can never have enough of :)