FAQ & troubleshooting

I get 401 even though I set a key. Check that the header name is exactly api-key, that there are no extra spaces or quotes around the value, and that you're not sending it as Authorization: Bearer ....

v2 returns 400 invalid_image for an image that works in v1. v2 requires standard base64. Common causes are URL-safe base64 (- and _ instead of + and /), truncated strings, and 16-bit PNG/TIFF files. Re-encode with a standard base64 encoder and save as an 8-bit JPEG or PNG.

The cat is obviously in the photo but I get not_detected or low_confidence. Try a closer, sharper, well-lit photo where the cat fills more of the frame. Rotate phone photos upright (EXIF) before sending them.

The bounding box is in the wrong place on my preview. Either the image wasn't rotated upright before upload (EXIF), or the box wasn't scaled from image.width/image.height to your preview size (Bounding box coordinates).

Should I send my threshold every time? Only if you need a specific operating point. Otherwise omit it to use the model default. prediction.threshold always tells you what was applied.

The results changed after a model release. /v2/predict follows the latest model. Pin /v2/models/bc-v1/predict if you need stable behaviour, and store model.id with each result.

Where do I report a problem? Contact support with the request_id, the time of the request, and the endpoint used (Support).