更新
每个 Lumenwerk 工具当前的版本，以及每次发布的变化。
Lumenwerk 正在开发中。低于 1.0 的版本为预览版；1.0 代表工具正式公开发布。
当前版本
发布说明
0.6.0
Autocut最新
Pick your AI model
Ten of the strongest AI models to choose from, right where you start the AI. Auto keeps picking the right one for each step.
- 新增A model menu: Claude, GPT, Gemini or Grok, grouped by smartest, strong and balanced, fast and cheap.
- 新增The price follows the model you pick and shows before every run, as always.
查看更多 · 还有 1 项变化收起
改进
- Auto stays the default: each step runs on the model that does it best.
0.5.0
Autocut
Tighter pauses, easier preview
Kept pauses stay short, silence hidden behind stretched words goes too, and the preview loops and scrubs.
- 改进Pauses Autocut keeps on purpose are never longer than 1.2 seconds.
- 修复Silence after a stretched word no longer slips through the cut.
查看更多 · 还有 2 项变化收起
新增
- After the cut, apply another tool straight away.
改进
- The preview starts again from the top when it ends, and you can scrub by dragging.
0.4.0
Autocut
Autocut on its own
Use Autocut without the editor: drop in a video, pick what to cut, run it and export. Or open it in the timeline as before.
- 新增Tools > Autocut asks how you want to use it: on its own or in the timeline.
- 新增On its own: upload, settings, run, then straight to the export (MP4, audio, or a project file for Resolve, Premiere and Final Cut).
- 新增Watch the result before you export: the preview plays the cut version (or the original to compare), and you can untick any cut right there.
查看更多 · 还有 1 项变化收起
新增
- Want to check the cuts? Open the result in the editor with one click.
0.3.0
Autocut
Cuts like an editor
After transcription, an AI reads the transcript like an editor and decides what goes. Autocut still places every cut on a clean point in the audio.
- 新增Retakes are found even when the second take is worded differently. The last complete take stays.
- 新增Abandoned sentences and remarks like "wait, again" are cut. Words like "basically" or "halt" only go where they add nothing.
- 改进Every AI cut shows why it was made. Cuts the AI did not confirm stay in the list, switched off.
查看更多 · 还有 1 项变化收起
改进
- The AI review is saved with the project, so reopening it costs nothing. Only the transcript text is sent, never audio or video.
0.2.0
Autocut
Natural pacing
Edits breathe like a person talks, and cut points get cleaner. Autocut now listens to where speech really starts and stops instead of trusting the transcript's timestamps alone.
- 改进Sentence ends keep a slightly longer pause (0.4 s) than joins mid-sentence (0.25 s), so the edit no longer sounds breathless.
- 改进Pause edges follow the real start and end of speech in the audio, also when the transcript's timestamps are off.
- 新增Coughs, throat clearing, sniffs and lip smacks are cut as their own kind. Laughter always stays.
查看更多 · 还有 7 项变化收起
新增
- Finds pauses the transcript hides, where a word's timestamp runs on into the silence after it.
- Option to keep the breath before a new sentence. By default the breath goes out completely instead of being cut in half.
- Fillers glued into running speech, where a cut would click, stay suggested but start switched off.
- Suggests abandoned sentence starts that break off mid-word, switched off by default.
改进
- More of the kept pause sits after a word than before the next one, so word endings and reverb tails aren't clipped.
修复
- A cough inside a pause no longer splits it into two pauses that are each too short to cut.
- "dass das" and similar word pairs are no longer flagged as stutters.
0.1.0
Autocut
Autocut is here
Autocut finds the parts of a recording nobody needs to hear and suggests them as cuts you review one by one.
- 新增Finds long pauses, filler words like "um" and "uh", stutters and retakes. For retakes, the last take stays.
- 新增Every cut can be switched on or off in a review list, with transcript and timeline side by side. Uncertain cuts start switched off.
- 新增Export as a project for Premiere Pro, DaVinci Resolve or Final Cut Pro (XML / FCPXML) plus SRT subtitles.
查看更多 · 还有 4 项变化收起
新增
- Pauses are shortened to a natural pause instead of a hard jump.
- Filler words in 90+ languages, plus optional soft fillers like "basically" or "halt".
- Cut edges snap to the quietest nearby frame, so no syllable gets clipped.
- The video stays on your computer. Only the extracted audio track is sent for transcription.