Detection
Recheck versus classifier: which Pro check to use
The classifier is fast, numeric and names a generator; the recheck reasons in words. Use the classifier first and the recheck for borderline cases.
Updated
Pro accounts have two paid checks for images. They answer different questions.
| Generator classifier | Model recheck | |
|---|---|---|
| Question it answers | Was this made by a generator, and which one? | Does this look generated, and why? |
| Output | Probability plus per-generator scores | Probability plus written notes |
| Runs | With an image detection batch, under the Pro classifier budget | Only when you choose Recheck with model |
| Daily limit (Pro) | 600 calls | 300 calls |
| Also for text | No | Yes |
| Shared | Cached for everyone | Cached for everyone |
How they combine
When both exist the score is 60% classifier and 40% recheck. Either alone stands on its own. A trusted provenance signal (Content Credentials, a watermark match, generator metadata) outranks both and sets the floor at 0.9.
Practical advice
- If the card already shows a Generator classifier row, that is usually the better signal for images; ask for a recheck when the probability sits near the middle or you want to see the reasoning.
- For text there is no classifier; the recheck is the only paid second opinion, and the writing-pattern notes remain the first.
- Neither is proof. See Why false positives happen.
Was this helpful?