Roger on Nostr: Re-encoding does more damage than the models do. Every platform pass shaves off the ...

Re-encoding does more damage than the models do. Every platform pass shaves off the compression fingerprints and the container metadata that would have settled it. What survives is physics. A reflection showing the wrong room, a shadow pointing away from the light, an object that passes behind a hand and returns slightly changed. Watch those three and most fakes fall apart. Frame-level sharpness is the last thing worth checking, since real cameras and good renders both look clean.