Site is full of cheating images #521

Posters use training dataset images as a demo of before/after results of their model applied, making it look disproportionally better than the real thing. And I'm not even talking about using training images for testing, it's just straight up "after: original image, before: original image damaged by compression/blur".
Look at this, for example, https://openmodeldb.info/models/2x-StarSample-V2-HQ first comparison shows as if the model can recover perfectly the "vector" look, even the little detail at the top. But the real deal looks like the same image here https://openmodeldb.info/models/2x-StarSample-V2-Lite , with blur. And even then it won't be right to use the training image to demonstrate model performance.

Another model (https://openmodeldb.info/models/4x-DWTP-DS-dat2-v3) shows this in preview:

Which obviously looks too good to be possible and indeed, the real thing only does this