Taan v1-preview scores music to your cuts, Riverside's Tonality edits a single spoken word, and HyperFrames 0.9.40

Taan v1-preview. Taan's founders (@mp3addi, YC) launched v1-preview on Nov 12 as a music model you can steer. Their pitch is that today's music models make catchy tracks but skip instructions like 'violin at exactly the 11th second', so you reroll until one fits. Taan takes a video, a script, reference audio, a hummed melody, MIDI or an image, reads the cuts and mood, and writes a score that pauses and shifts on the cuts. You can then change one instrument or one section (move the violin from 8 s to 5 s, say) while the rest stays put. Per the company, outputs are original with commercial rights, it was trained on 100% licensed music, and you can download stems. v1-preview covers orchestral and cinematic only for now. The launch post passed 340 likes and 100k views in its first few hours.

Why video people should care. The useful part is the MCP and API access. Wire Taan into Claude Code or Codex next to a Remotion, HyperFrames or raea project and the agent can request a score timed to its own cut list instead of you dragging a stock track around. @maxxmalist flagged it for AI ads and short-form work 'where every second matters'. Plans start at $15 a month for 25 minutes of music (Creator) and $45 for 60 minutes (Agency), both with MCP/API access. Those are the site's list prices. Accuracy on timing is the company's claim; we haven't seen independent comparisons yet.

Riverside Tonality. Riverside Labs' Tonality is a voice model that edits words inside a real recording. You can change, insert or remove a word, or add a sentence, and set the pitch, tone, duration, pause, energy and pronunciation of each word. Per Riverside, the speaker's voice, accent and room acoustics stay as recorded, where most voice tools regenerate the whole sentence. There's a free, rate-limited playground and API access through sales. It's the fix-a-flubbed-line tool for voiceovers and talking-head cuts. One open question from replies, raised by @maxxmalist himself, is how it handles proper nouns. Use it on your own recordings or with the speaker's consent.

HyperFrames 0.9.40 (Nov 12). A local render now extracts each source video at about the size it's shown, so a long export of 4K footage inside a 1080p film no longer fills the disk. A frame's pixels no longer depend on frames drawn before it, which should make renders more repeatable when you re-render a range. Render work is sized to the CPUs the process is allowed to use (useful in containers), 'check' now handles WAV files whose header misstates the byte rate, and 'hyperframes registry search' runs the catalog. The Docker image moved to Node 22.23.5 for November's security release.

Signals: Taan and Tonality are official launches amplified by tracked creator @maxxmalist; HyperFrames is official release notes. No tracked curator pick or GitHub Trending signal, so nothing here is HOT by our rule.