Linesmith 4.6.0

Linesmith 4.6.0 makes dubbing use far fewer requests and brings the new voice closer to the original performance.

• One request for a whole scene: Gemini now reads the whole script in one request, including two-speaker scenes. Extra takes are read inside that same request, so 3 takes is now the default for Gemini and costs no more requests. A line is no longer asked for again just because it ran long.
• Lines are found on your computer: Linesmith finds every line in the generated speech with a word-timing model for the dubbing language. The first dub into a language downloads that model once (about 1.3 GB for Turkish).
• Timed to the lips: each phrase is placed where the original speaker's mouth moves and gently stretched to fit, including whispered lines.
• Closer emotion: the acting notes name how loud each line is and describe the scene, so the voice plays the characters. Quiet, breathy lines are no longer treated as shouts.
• Reading check: a line that was skipped, or read with words added or left out, is noticed and asked for once more on its own.
• ElevenLabs, OpenAI and Microsoft Azure also dub a whole script in one request: ElevenLabs v3/v4 read every voice in one dialogue request with short audio tags (Eleven v4 is now in the model list), OpenAI gets each line's direction, and Azure reads every line in one request with its time and a speaking style when the voice has one. On these services extra takes still use characters.
• Privacy: when another service speaks the dub and a Gemini key is set, Gemini can listen to the original speech to write acting notes. Linesmith asks before sending it, and the privacy notice (version 2026-10-05.2) describes this. Because the notice changed, you will see the terms screen once more.

Updating: Linesmith 4.4 and 4.5 update themselves to 4.6 when you close the app. From 4.3 or earlier, download and run this release once.

Downloads and SHA-256 checksums are attached below.