My contribution and what reached Beta

I owned the end-to-end interaction design for Generate Soundscape: source selection, generation states, track placement, editing, and recovery. I carried those decisions through design reviews, final specifications, and engineering handoff.

When this work began, Generate Sound Effects was the only generative audio tool available in Premiere. Soundscape introduced a more complex workflow: creating sound effects and ambience together, placing multiple layers, and letting you edit or regenerate each one. I worked with the Gen Audio team to extend the existing experience for that new capability.

The August 2026 Beta produces four separate audio layers: two background and two foreground. You can move, mix, and regenerate each one. The initial release supported single scenes of roughly 15 seconds or less.

Make clear which video Premiere will interpret

A selected clip and a timeline range can point to different footage. I defined both entry points so you can understand the source before spending credits: one clip, or the visible picture across a range. Hidden or covered footage is excluded. Team review surfaced a risk: a generation outline on selected video could imply overwriting it. I kept that cue off the video and made the source explicit in the controls.

Add new layers without disturbing your existing edit

Early exploration asked you to assign outputs to numbered tracks in a separate step. That meant translating those labels back into a timeline you could already see. I proposed placing the layers directly in the timeline, where you could move them yourself.

An interim design pushed existing tracks down to make room. The final specification instead keeps new layers together on available tracks, skips occupied tracks, and creates tracks when needed. During engineering review, we clarified that locked tracks count as occupied too. I incorporated that rule so automatic placement respects the boundaries of your existing edit.

Generate a starting point you can keep shaping

You can rebalance the background or regenerate one foreground sound without discarding the whole soundscape. I adapted the existing Sound Effects panel with the Gen Audio team, giving you a familiar place to refine the new multi-layer outputs. In the specification, Add Timing lets you record your own timing. Sync to Video matches the sound to the footage instead. I required you to remove a timing recording before enabling automatic sync, so it would not silently replace your work.

I avoided promising an exact output count while the model was evolving. The Beta settled on four layers, each editable on its own.

A failed attempt should leave an understandable timeline

You can move footage while generation runs. Engineering review exposed a retry problem: a failed placeholder might no longer refer to the same picture. I explored returning you to the source selection. For the first implementation, we retained the shared workspace’s failed-clip behavior: label the failure, then start a fresh attempt from footage. The final specification places retries on available tracks without overwriting the failed attempt.

Final specification: visible failure, then a fresh attempt from your footage.