can you explain what you mean by 2 steps without lora? You mean like having 3 samplers?

Howdy, when I download the newly uploaded "T2V" the file is actually labelled as "I2V".

A question, can I use this lora and Pusa V1 also with Fusion X or do I need to have the specific model you link? I use WAN 2.1 for now. I noticed that with Fusion X if I set the parameters you indicate for T2V the characters' skin has a lot of glare. How can I improve this condition, which I don't like very much?

Coming back to say thanks for the new loras and also the information about the tripleKsampler. That sampler has made WAN22 much easier to use and the quality is MUCH better. Whoever came up with that is a godsend!

Hey is the newest one different than this ?

https://huggingface.co/Kijai/WanVideo_comfy/tree/main/LoRAs/Wan22_Lightx2v

I feel like such a noob, can someone link me a workflow that uses the tripleKsampler?

Successful run of Triple K Sampler on 3060 w/12gb, looks very decent from a 640x368 gen to FHD topaz upscale, 8 minutes for 49f!

For some reason i keep getting lora key not loaded with 1022 and then the output comes extremely noisy, i read the documentation on the huggingface page and it says we need clip vision hooked? i thought clip vision was no longer needed for Wan 2.2?

My final verdict :

Using the 1022 high LoRA at 0.5–0.75 strength only on lightx2v 4-step distilled high noise model, gives me the results I want most of the time. The motion appears more dynamic, and prompt adherence is slightly improves when the 1022 LoRA is used alongside the Lightx2v 4-step distilled model.

I also tested the 1022 LoRA with the base WAN 2.2 model, but the results were not as i expected, But it's still good tho if you need less motion.

I'm still using 2 ksampler not triple. Using tripleksampler takes too long on my machine. So i go back using 2 sampler instead cuz it gives me reasonable gen time and quality.

the Lightx2v 1030 GGUF distiled model :

https://huggingface.co/jayn7/WAN2.2-I2V_A14B-DISTILL-LIGHTX2V-4STEP-GGUF/tree/main/high_noise_1030

*Only use the high 1022 LoRA on the high distiled model only and don't crank up the lora strength to 1 if you’re using the distilled model,

the distilled low model already includes the low LoRA. The included LoRA is still the same one from Lightx2v WAN 2.1, so it’s not a big deal. If you add the 1022 low LoRA to the low distiled model, it’ll just increase generation time without much improvement, and if you use lora strength above 0.8 on high distiled model, the consistency (ie. face ) would start to change from the original image.

Which Lora high or low is compatible with Wan 2.1 bcoz wan2.1 doesn't have low and high

In the new 1022 version, I’m getting a “LoRA key not loaded” error.
Is there any way to fix this?

How do you guys avoid the prompted action being interrupted after just a couple seconds then returning to the initial point as if it tried to make a loop?

Using WAN2.2-I2V_A14B-DISTILL-LIGHTX2V-4STEP-GGUF
with the high noise version of this 1022 LORA at .7 + other LORAs combined, length 169 with 24 FPS cfg 1 steps 4 euler simple in both samplers.