Vision Finetuning failure: Too many images were skipped due to moderation

API
1

I tried to use the new vision finetuning feature but ran into this problem.

The job failed due to an invalid training file. Too many images were skipped due to moderation. Please ensure that your images do not contain content that violates our usage policy.

It’s hard for me to check which images might be problematic. They are all cartoon (Pokemon) images.
I’m wondering whether someone can help unblock me. I’m happy to provide job id and training data.
Or, it would be nice if I know which of the lines were causing this problem, and what kind of violation it was.

1 Like
2

Maybe reading the usage policy could give some insight?

Seeing as you said Pokemon, I wonder if its flagging for copyright reasons

1 Like
3

Thanks. I don’t see the policy explicitly mention images with copyrighted characters cannot be used in fine tuning.
I hope an OpenAI employee can verify if this is the case.

4

Maybe try asking on this thread, it was created by an OpenAI employee