Movie still of then the rest of your prompt. Your mileage may vary if you're using lightning loras. Try between 0.4 and 2.0, 0.8 is usually ideal.
It works well with loras like studio photo style and lenovo at low weights.
Description
FAQ
Comments (5)
Why does Qwen always look so burned?
It's just the way it is if you use lightning loras. If you use the lightning loras, try a two step process of the first 4 steps without the lightning lora, or it at a low weight, then the rest of the steps using the lightning lora, it usually works out less burned in.
what sampler / scheduler are you using? I have tried a few and some look better. Currently using res2s Bong tangent and that isn't too bad (at least doesn't have the halo) but I've not tried that many).
@EricRollei21 I've just been using euler/simple or euler bong_tangent. res_2s is my other go-to, but it's very slow as it's two steps in one, and qwen is already slow enough.
Throw in a CFGNorm node and set to around 0.92-0.96.