1.0.2 — the clips get their words back

Every clip now leaves with a line written about it, in the language of the recording.

Before this update that only really worked in Russian. An English session took its wording from a fixed phrase library, and a session in any other language got a sentence lifted straight out of what was said, unpunctuated, exactly as it came off the transcript. Several of the checks that exist to stop a weak line from going out could only read Russian, and one of them was throwing away perfectly good English captions for the crime of ending in a full stop.

That is what changed. The line is written for each clip, checked in the language it was written in, and it goes out in that language or not at all.

Two of the ways CutBench reads a session were not running

One of them reads what the game itself prints on the screen. The other follows what the player's voice is doing: a laugh, a shout going up, the short silence before something goes wrong.

Both of them shipped without the files they needed, so both quietly produced nothing, in every release so far. The run still finished and the clips still came out, which is exactly why this went unnoticed for so long.

They work now. On a session with a lot of banter, or a lot happening on screen, the picks should be visibly better than they were last week.

The first second

A clip is now judged on its own opening frame as well as on the moment it was cut for. The thumbnail is that frame rather than one from somewhere in the middle, and the list is ordered by how well each clip holds attention in its first second.

Smaller things in this build

• The YouTube title sits on the clip card, pre-filled and counted, so nothing is added behind your back at upload time.
• A long session no longer loses its written copy. The pass that writes it used to be cut short on recordings with many clips, and the wording fell back to templates without saying why.
• The install is about 3 GB smaller.

About graphics cards

Cutting, ranking, rendering and burned-in subtitles work from the minimum card listed on the store page.

The written titles and descriptions come from a model that runs on your own machine, and it wants 10 GB of video memory. Below that the clips are still cut, ranked and rendered, and the wording falls back to templates. The run notes say which case you are in rather than leaving you guessing.

As always: nothing is uploaded, nothing is transcoded by a stranger, and the licence checks once and then leaves you alone.