Proof, not promises

Don't approve a generation because of the model name or the prompt behind it. Approve it because of what you can actually see: what changed, what stayed the same, and what's broken.

Judge the result, not the model

These three frames are real generations from the team. For each one, decide: does it hold up, or does it need another pass? Base that only on what's in the image — a strong model can still produce a result that isn't ready.

A model's reputation can tell you what to try. Only the image in front of you can tell you if it worked.

Same edit, different verdicts

The note calls this next result a "Nano Banana Pro one-pass re-dress." Ignore that — judge the pixels. Toggle between the two sheets below and check three things: does the horned identity hold, do all three views (front/back/portrait) still agree, and is the new armor consistent everywhere?

So the verdict depends on what happens next: proceed if you just need an armor design. Hold if this needs to carry the creature's face into Seedance — the face is no longer visible, so that part still needs fixing. Same image, two different answers, because the downstream use is different.

The same test, on a new edit

Same method, a different edit from the same source. A pass on the last one tells you nothing about this one — check it fresh.

The face and body hold up. The new blade-arms show up in both views, but their dark edges blend into the black background the same way the armor did — worth fixing before animation.

Watch for what changed that you didn't ask for

The note says "Seedream 4.5 added wings and blood cleanly." Ignore the word cleanly — that's a claim, not a fact. Check every part of the sheet, not just the new wings.