<< Click to Display Table of Contents >> Navigation:  »No topics above this level« On-Screen text collisions

On-screen text collision detection analyzes the video picture to identify areas occupied by signs, credits, annotations, and other burnt-in text. This information is used during Quality Control to flag subtitles that may obscure or visually clash with existing text on screen. By identifying these overlaps, you can make necessary adjustments to ensure both the subtitles and the original video text remain clear and readable.

Requirements and Performance

On-screen text collision detection runs locally on your computer and requires the installation of an additional tool: the EZTitles AI Features Pack. It uses OpenCV libraries and the EAST (Efficient and Accurate Scene Text Detector) deep learning model to detect burnt-in text and determine the areas it occupies, represented by rotated bounding boxes and quadrilaterals.

The on-screen collision detection starts automatically when a video is opened, imported. If you're working with EVS videos, on-screen collision detection happens during ingest and cannot be turned on/off in addition afterward.

Because video processing is resource-intensive, performance depends on your computer's hardware, and analysis may take longer on older systems.

The feature is turned on by default, but you can turn it off at any time by navigating to Edit > Preferences > Performance.

Checking Subtitles Against On-Screen Text

You can customize your QC checks from Check Subtitles > On-screen text to specify which types of overlaps should be flagged. They help identify the subtitles that may obscure existing text, either by occupying the same screen region, overlapping its vertical position, or directly covering the detected text area. This helps spot subtitles that conflict with on-screen text—whether they share the same general region, overlap in vertical alignment, or directly cover or intersect the detected text area.

Adding Video Mask excludes a specific region from collision detection. This is useful for areas containing burnt-in text that is consistently flagged as a source of collisions, such as running timecode, burnt-in subtitles, or logos. These elements may be omitted from the final, production-ready video or may otherwise be irrelevant to subtitle placement. On-screen text within a masked area will not be considered a source of collisions during QC.