Turn a product changelog into a 30-second update video

To turn a changelog into a 30-second video, keep three changes, write about 70 words, generate the voiceover with Sume TTS, drop one real screenshot per change into a Timeline 1.0 render and let the voice length set the cut points. The machine cost is about $0.12: roughly $0.02 of TTS plus $0.10 for one rendered minute. You supply the screenshots.

Software teams ship notes faster than they ship video, so the useful habit is a template you re-run every release.

Pick three, not ten

Changelogs list everything. A 30-second video can carry three beats. As a worked example, take Raycast's v2.6 notes, published 30 September 2026: the host process uses about 50% less memory, extension processes use about 30% less memory at startup, and the Auto model has Lightweight, Balanced and Flagship preferences, with Balanced using about 70% fewer credits than Flagship. Those are three claims with numbers, which is what a short video can hold. Quote the vendor's wording exactly; do not round or embellish a number you read.

The pipeline

Cut on the words

Ask TTS for word timestamps and place each screenshot's start at the first word of its sentence. The completed TTS result carries monotonic words[] with start and end seconds, so you do not guess. Put the TTS file on audio.url with its measured duration, give each slot fit: "contain" or "blur" so a desktop screenshot is not cropped in a vertical frame, and add a 0.25-second fade between slots. If you want phone-sized text, burn a caption pass afterward at $0.20.

Use real screens, not generated ones

An image model can draw a plausible app window, but it will not match your product. Take screenshots from the product, or pull stills from a screen recording with POST /v1/video-frames, which is unbilled. Generated imagery belongs in the title card, not in the proof.

Limits

The script is yours. Sume does not verify that a changelog claim is true or still current, so link the vendor's notes in the post and re-read them on the day you publish. TTS lengths vary by voice, so measure the output, then trim the script rather than speeding the audio. A single TTS job takes up to 20,000 characters, far more than this needs.