Anti-Scam OCR setup
Protection against fake giveaways, phishing screenshots and lookalike DMs — 100% local.
What this does
SlakBot's Anti-Scam scans every image posted in your server. Using local OCR (no external AI), it detects:
- Fake Nitro screenshots — "Free Nitro from MrBeast" etc.
- Phishing Discord-UI clones — fake "verify here" prompts
- Lookalike DM screenshots — accounts impersonating admins/mods
- Crypto/giveaway scams — known scam-DM patterns
How is it different from Discord AutoMod?
Discord's AutoMod only works on text. Scammers bypass it by posting their fake content as screenshots — AutoMod can't read those. SlakBot's OCR can.
1Activate Anti-Scam
Open the dashboard. Navigate to Security → Anti-Scam. Toggle on.
That's enough to be protected — the default settings are fine for most servers.
2Pick action on detection
Three modes:
- Log only — writes to mod-log, takes no action. Good for testing.
- Delete message + warn — auto-delete, ping mod-log, give the user a /warn. Recommended default.
- Auto-ban — aggressive. Direct ban on detection. Only for servers with many scam attacks.
3Configure mod-log channel
Select a mod-only channel (e.g. #anti-scam-log) where every detection gets logged. Example output:
Type: fake-nitro screenshot · Confidence: 94%
Action: deleted + /warn
4Whitelist (optional)
Some content might be a false positive (e.g. a legit MrBeast collab post). Whitelist:
- Per role — e.g. @Mods are always whitelisted
- Per channel — e.g.
#meme-screenshotsdoesn't get scanned - Per user account age — accounts > 1 year old get a lower threshold
5Fine-tune (advanced)
In Security → Anti-Scam → Advanced you can:
- Confidence threshold (default 75% — lower for aggressive, higher for fewer false positives)
- Sensitivity per type — e.g. Nitro detection high, crypto scams lower
- Pre-emptive ban — if an account is < 1 day old and posts a scam screenshot: instant ban (default off)
False positives?
If a legit message is wrongly flagged:
- Mods can restore the message via
/restore-message - Add the user to a whitelist role
- Or send the screenshot to @luca_046 — we continually retrain the model