Four things happen after you hit create. Nothing else.
Your script is read aloud by a text-to-speech engine. Four voice styles are selectable, and the final video length follows the generated audio rather than a number you picked, so the last word is never chopped off.
Captions are generated per word and rendered as a real subtitle file, then burned into the picture. Four styles: solid yellow, white with a yellow highlight, a word-pop scale effect, and a plain bottom caption.
A random segment is picked from the full-length footage for the game you chose, so two videos never open on the same moment. Segment selection avoids the last timestamp it used.
Gameplay is cropped to vertical, scaled to 720×1280, and encoded to H.264 with the narration mixed in and captions burned in. No watermark is added.
Large Impact text with a heavy black outline. The default, and the most readable at small sizes.
White text where each word lights up in turn as it is spoken.
One word at a time, scaling up briefly before settling. Suited to fast narration.
Small white text at the bottom of the frame with a soft shadow. No outline.
No music library, no background music, no effects or transitions, and no background removal. Captions are timed from a character-weighted estimate of the script rather than a speech-recognition pass, so on very long scripts the timing can drift slightly from the audio.