Waveform editing

• They edit this node in place: the result goes back onto the same card, and the edges out of it, the node's name and its place in a group all stay put.
• No model, no credits: the whole clip is re-encoded in your browser, so there is no queue to wait in either.
• ⌘Z takes it back: a copy is kept before the edit, so undo returns you to the clip you had.

Hit Trim on the toolbar and a selection appears on the waveform, with a bar opening along the bottom edge of the card.

• Drag the left handle to move the start, the right handle to move the end, and the box itself to slide the whole selection along (its length does not change).
• As you drag, a bubble reading x.xx s floats above the box telling you how much you have selected. The shortest selection is 0.2 seconds.
• The bar below reads 00:00 - 00:05 on the left and has the round ↑ button on the right — click it and everything outside the selection is cut away.

Hit Speed on the toolbar and the bar along the bottom edge of the card turns into a slider.

• The range is 0.1x ~ 4.0x, 1x by default. The slider moves in steps of 0.05, so to land on an exact rate, type it into the number box on the right.
• Speed applies to the whole clip; you do not have to select a range first.
• Once you confirm, the duration on the card follows (faster is shorter, slower is longer).

Trim and Speed edit this clip itself; the toolbar also has two buttons that split one clip into several new ones — the source node stays untouched, and each segment lands as a new audio card (each with a wire back to the source):

• Smart split — one click, no questions asked: it cuts by phrase (the breaths and pauses between sung or spoken lines), then snaps each cut to the nearest beat — every segment starts on the beat, ready for video cuts and rhythm sync. A full song or a finished narration usually comes out right with this first pass.
• Custom split — when that misses, say it yourself: "keep 0-2s", "delete 10-15s", "cut at 5s and 12s", "keep 0-3s and 8-12s". Clicking drops a task card to the right; write your cuts in the bar beneath it, the sentence is turned into "which spans to keep", and the pieces are cut one by one.

Both are free: only the sentence passes through a text model once — the actual cutting and encoding happens in your browser, with no queue.