Wan-Lighting : 4steps per model or 4steps total?
Hi
should I use 4 steps for the high noise model, plus 4 other steps for the low noise model (so 8 steps total) ?
or 2 for each one (4 steps total) ?
Considering how it makes it so fast, I don't mind waiting a bit more to have high quality
But I'm thinking it may be a bad idea to let the high noise model run too much (it would denoise too much, while that model is designed to be good when it's noisy...)
Thank you Kijai !!
It's 4 in total. Please check this workflow https://huggingface.co/lightx2v/Wan2.2-Lightning/blob/main/Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1/Wan2.2-I2V-A14B-4steps-lora-rank64-Seko-V1-forKJ.json .
Wow thanks.
If I want to increase quality (a bit sharper skin mainly) should I increase low noise (second pass) to 6 or 8, but not touch the first high noise pass ?
Wow thanks.
If I want to increase quality (a bit sharper skin mainly) should I increase low noise (second pass) to 6 or 8, but not touch the first high noise pass ?
less high model steps and more low model steps will focus on image quality
higher shift will focus on composition and movement while lower shift will focus on appearance and texture
Thanks, make sense.
Though using less of the high noise model also means getting less of the fancy wan2.2 improvements (as the low noise model is just wan2.1 finetunedy while the high moise model is brand new)
I guess doing 2 steps with high and 6 steps with low would be nice hmm