ClipSonic

Changelog

Stay up to date with the latest improvements and new features

v1.0.1 Latest
  • New: Smart Viral Clips — a third way to find your best moments. It scores every part of your video on pacing, hooks, pauses, and energy to pick the strongest clips, with no AI model involved. Pick a target clip length and how many clips you want, or leave it on Auto.
  • New: choose your clipping mode right at project start — AI Viral Clips (AI finds and ranks your best moments), Smart Viral Clips (rule-based, no AI), or Fixed-Length Clips (auto-splits the whole video at a length you pick).
  • Renamed: "Smart manual clips" is now Fixed-Length Clips, so it's not confused with the new Smart Viral Clips. How it works hasn't changed.
  • Fixed: face tracking now picks the right person to follow. In videos with several people on screen, it could lock onto someone in the background instead of the main speaker — it now looks across the whole clip first and follows whoever's on screen the most, most prominent, and most centered. This also improves the two-person split-screen layout.
v0.8.0
  • Added support for multiple Groq and Google AI Studio API keys, used in order with automatic switching when one hits its rate or daily limit.
  • Added the same automatic key switching to transcription, which now shares your saved Groq keys.
  • Improved long videos: highlight detection keeps the clips it already found when a provider limit is reached, instead of discarding the whole run.
  • Added a Resume button that continues highlight detection from where it stopped, without re-spending quota on parts already analyzed.
  • Added a notice in the Clips stage showing exactly how far analysis got, for example "stopped at 1:00:00 of 4:00:00".
  • Improved transcription of long audio, which now resumes from the last completed chunk rather than re-uploading everything.
  • Improved reliability on long videos: a single failed section is skipped and reported instead of ending the whole analysis.
  • Improved "Test connection", which now checks every saved key and reports which ones work.
  • Updated the Groq model list: Llama 3.1 and 3.3 were retired by Groq, so Qwen 3.6 27B is now the default. Saved settings pointing at a retired model move to the current default automatically.
v0.7.0
  • Added the ability to edit one video setting and apply it to all videos.
  • Added the option to save videos with AI-suggested titles.
  • Improved several core features for better performance.
  • Fixed minor bugs and stability issues.
v0.5.0
  • Fixed Open AI Issues
  • Added Google Gemini as LLM provider.
v0.3.0
  • Added a new 1+ Minute Videos toggle to recommend longer videos for better clip generation, along with major performance improvements, UI enhancements, and multiple bug fixes.
v0.2.1
  • Fixed the Groq free-tier "Request too large" (413) on highlight detection.
v0.2.0
  • Added: Qwen 3.6 27B (131k context) to the Groq model catalog for highlight detection, in both the Rust provider list and the frontend model picker.
  • Added: Custom YouTube cookies file support for yt-dlp, enabling downloads of age-restricted, members-only, region-locked, and bot-gated videos.
  • Added: `youtube_cookies_path` app setting, persisted in settings.json and passed to yt-dlp as `--cookies` for both metadata probes and downloads.
  • Added: yt-dlp cookies are applied only when the configured file exists on disk, falling back to anonymous requests when unset or stale.
  • Added: "YouTube cookies" Settings card with an attach/change/remove file picker, a security notice, and a Cookie-Editor export guide linking to the extension.
  • Added: Persistent guidance on the New Project screen pointing users to attach a cookies file if a video fails to fetch or download.
  • Added: Contextual notice on New Project that surfaces the cookies-file fix when a download error looks like a YouTube sign-in/bot gate.
  • Improved: Highlight detection now scans the transcript in overlapping 10-minute windows so moments spanning a window boundary are no longer missed.
  • Improved: Out-of-band clip proposals are fitted into the 10–120s range — over-long clips trimmed, short-but-plausible clips grown — instead of being discarded.
  • Improved: Short source videos receive a scaled-down minimum clip length so brief uploads still yield usable clips instead of analyzing to nothing.
  • Improved: Near-duplicate suggestions from overlapping windows are de-duplicated, keeping the higher-scoring clip.
  • Fixed: Empty analysis results now return an actionable error instead of silently leaving the "Find highlights" screen unchanged.
v0.1.0
  • Minor Issues Fixed
  • changed hard clip duration to a minimum of 20 sec