← Blog

Text moderation, not just images — on our own model, in Polish

2026-07-02 · admin@modwall.dev · approx. 1 min read

Inappropriate content isn't only photos. Toxic comments, threats, insults and text spam hit a community just as hard as NSFW in an image — and under the DSA you are responsible for both. That is why modwall now moderates text as well.

The same pattern, a second endpoint

You check text via /v1/moderate-text — exactly like images: a raw toxicity score, the same monitor / review / enforce modes, the same HITL queue and audit log. Integration is a few lines.

curl -X POST https://modwall.dev/v1/moderate-text \
  -H "Authorization: Bearer sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{"text":"treść do sprawdzenia"}'

Multilingual — it understands Polish

The model is multilingual, so it catches toxicity in Polish too, not only in English. That matters: an English-only classifier would let most Polish abuse through.

Our own model, no GPU, no dependencies

The text model runs on our side (or on yours in self-host) on CPU — no external API that will change its pricing or shut down its free plan tomorrow. We process the content in memory and do not store it; we log only metadata.

Pay only for what you use

You choose the scope: images only, text only or both bundled (cheaper than separately). One shared request limit. A forum mostly needs text, a marketplace mostly images — everyone pays for their own.

Create an account and choose your scope, or take a look at the API documentation.