Changelog

Shipping weekly.

Cutfarmer is developed in the open with the creators who use it daily. Highlights from recent beta builds:

Beta August 2026

Instagram Reels publishing + hook titles everywhere

  • Publish finished clips to Instagram Reels through the same pipeline as TikTok and YouTube, with daily-quota awareness and automatic cover-frame selection (the biggest, mid-speech face — no more blinking thumbnails).
  • Hook titles now burn on every clip by default, auto-positioned in the top band so they never cover the speaker or the watermark.
  • Title wording safety: measured across live posts, explicitly-worded titles never reached the feed — the publisher now diverts them for a rewrite automatically.
  • Cross-platform learning: yesterday's Reels performance ranks today's upload queues.
Beta July 2026

Publishing quality gates + smarter clip boundaries

  • Every render now passes a coverage check (is a face in frame whenever someone speaks?) and a vision review before it can auto-post; anything questionable is flagged for human review instead of shipped.
  • The clip refiner re-cuts both boundaries per clip — hook-forward opens, payoff-complete endings — validated in paired A/B against original cuts.
  • Natural-length clipping: briefs like "one clip per comedian, the full set" are followed instead of chopping to a fixed duration.
  • Upload budgeting: per-channel rolling 24-hour caps surfaced in the publish queue so you never hit a platform limit blind.
Beta June 2026

Identity-aware framing engine

  • New face-identity backend recognizes the same person across camera angles and lighting — the crop follows the comedian, not the biggest face.
  • Voice-to-face matching keeps framing correct on crowded panels, through laugh breaks and walk-ons.
  • Motion-aware holds: broadcast-style steady framing with slow follows and hard cuts where an editor would cut.
  • Channel subscriptions: follow a show and every new episode is ingested and clipped automatically on a schedule.
Beta May 2026

Native render pipeline

  • Rendering moved fully to AVFoundation — hardware-accelerated, color-managed 1080×1920 output with loudness normalization.
  • Preview is now the export pipeline at a smaller bitrate: what you see in the canvas is exactly what ships.
  • Word-level caption engine with per-word highlight timing and template-driven styles.
  • On-device transcription and speaker separation, several times faster than real time on Apple Silicon.

Earlier builds covered the original three-phase workspace, the ingest pipeline, and the first framing engine. Full history ships with the app.