GIF Creator Learn

Docs

AI blur detection

Use a vision-capable model to detect sensitive content and generate blur regions automatically.

Overview

AI blur detection is the product’s clearest differentiator. It turns a manual review task into a guided verification task by suggesting blur regions for sensitive content in a selected frame.

What the feature does

  • Sends the current frame to a vision-capable model
  • Creates blur regions for detected content
  • Labels AI-created regions with a PII classification, such as email address or phone number, so you can review them quickly
  • Tracks token usage for the current call and the full project
AI token usage banner in the editor
After a `GPT-4.1` AI run, the editor shows both per-call token usage and cumulative project totals.

PII categories the AI looks for

The default redaction prompt asks the model to find categories such as:

  • full names, usernames, and display names
  • email addresses
  • phone numbers
  • IP addresses, MAC addresses, and hostnames tied to real machines
  • passwords, tokens, API keys, secrets, and credentials
  • financial data linked to a person
  • passport numbers, national IDs, and other government-issued identifiers
  • internal system names, device names, and organisation-specific identifiers

The project README confirms these vision-capable models are suitable:

  • gpt-4o
  • gpt-4.1
  • gpt-4.1-mini
  • gpt-4.1-nano

Review checklist

After an AI run:

  1. Confirm each suggested region covers the sensitive content fully.
  2. Check the label and source badge.
  3. Adjust or delete incorrect regions.
  4. Propagate only the regions that stay valid across following frames.

Why teams like this feature

It compresses the slowest part of screen-recording redaction into one action, then keeps the human in control for verification.