Obelisk Image Spam Detection Premium
Obelisk compares pictures people post with a library of known spam and ad images, and removes matches.
How it works
Obelisk fingerprints images that are already known to be spam using computer-vision features and a perceptual hash. Each new upload is compared against a shared built-in library plus your server’s own vault. When the match confidence passes your server’s threshold, the message is deleted and the poster is punished.
This is image matching, not an AI classifier. It catches re-posts of known spam pictures; it does not judge arbitrary images.
Your server’s image vault
Add your own spam examples with /image_add. Each server has its own private vault: other servers can’t list, preview, match or delete your images.
- Accepted formats: PNG, JPEG, WebP and GIF (only the first frame of a GIF is used). The file’s real content is checked, not just its extension.
- Every image is re-encoded on upload, which strips metadata such as EXIF.
- Maximum file size is 8 MB.
- Free and Premium have different vault storage quotas and upload rate limits (Free: 25 MB; Premium: 250 MB).
Testing an image
/check_image checks an image URL and shows whether it matches known spam, the confidence and the matched image.
Settings
| Setting | Default |
|---|---|
| Similarity threshold (how close a match must be) | 65% |
| Delete matching images automatically | On |
| Exempt channels / always-scan channels | None |
| Detection rate limit | 10 per minute |
What happens to the poster is set under Punishments.