Who got close?
Ten drawings per model.
Automatic likeness: out of 100. Human pass rate: Bufo labels out of ten.
Waiting for human review.
| Rank / model | Human / auto scores | Cost |
|---|
About these prices
“Measured run” is what the API charged us. “Estimated” uses saved samples. A missing price stays missing. Prices keep their source label.
“Good value” marks a non-dominated cost / pass-rate option among models with a complete human review. It is descriptive for this ten-sample run, not a promise about every generation.
Missing review is not a zero. How the scores work
Pick your frog.
Both got the same prompt. Pick the more convincing Bufo, then see who made it.
Your picks stay in this browser. They don’t change the leaderboard.
The family resemblance varies.
How do the scores work?
Bufo can hold a coffee. Bufo can be the coffee. The good ones feel like someone had a joke and opened an image editor. The familiar face helps, but a rough redraw can work too.
Each model gets the same ten prompts and the original image. People label each returned image Bufo, Close, or Not Bufo. A Bufo label is one passing point; Close and Not Bufo are zero, with no half credit. Unreviewed or unattempted attempts prevent ranking. Verified generation failures count as zero points.
Read exact prompts and download the Bufo Maker skill