Thumbnail detection AI is not working correctly and wrongly flagging our thumbs

Summary

They mentioned on twitter this has been resolved.
We are running in countless issues still and the calibration doesn’t make sense.

Wrongly closed post:

Please select what you are reporting on:

Unreal Editor for Fortnite

What Type of Bug are you experiencing?

Publishing

Steps to Reproduce

  1. Upload a thumbnail

Expected Result

Work

Observed Result

Rejected for similarity reasons

Platform(s)

pc

Below variant is rejected, because of the banana peely soccer player. If we remove the soccer player entirely, it passes the check.

Below variant is accepted, since it’s no banana peely skin

We tried the following ‘kick’ character replacements as well, but they all get rejected.
We basically can not use ANY reference to the fishstick skin / banana peely skin in our thumbnails anymore. Which is definitely a calibration issue with the detection AI.

1 Like

@Flak another example of faulty calibration

REJECTED:

APPROVED:

1 Like

Same here! I made a map, and when I uploaded the thumbnail for the first time, it worked. I wanted to slightly tweak the thumbnail by making it brighter, and I got the error “thumbnail is too similar to another player’s island.” Epic doesn’t realize it’s MY thumbnail! Please fix the AI Epic. Does everyone notice this all happened around 25 days ago? It’s that new update. Has anyone found a solution to the issue? :face_exhaling:

2 Likes

Another example of us having to upload a thumbnail which has a lower CTR because of the wrongly calibrated AI detection.

‘To similar’

Passes validation:

I feel there’s a disconnect here, as the teams building these systems may not experience the publishing process from a creator’s perspective.

We’ve been developing full time in UEFN for almost 2 years and publish maps regularly. Happy to share feedback. Feel free to reach out.

1 Like

Another example:

Passes validation:

Fails validation:

There’s literally no one with a water pistol on his thumbnail like this.

@Flak every thumbnail we try to upload we have a lot of effort trying to make it pass.

The one which now fails validation is way better than 1st one which passes. Now we’re forced to tweak it more.

The calibration 100% makes no sense.

1 Like

(post deleted by author)

FORT-1116255 has been created and its status is ‘Unconfirmed’. This is now in a queue to be reproduced and confirmed.

few hours later, this passes validation:

But if we want any rainbow text, it fails.

Same if we do rainbow text above the elephant.

I can’t believe that this is intended behavior of the AI calibration.