Adobe Firefly can turn your vocal cues into custom sound effects

Most AI video generators still spit out silent films. If you want footsteps or a door slam, you are usually stuck hunting through stock audio libraries or manually nudging waveforms on a timeline.

Adobe is trying to change that with a surprisingly fun shortcut. On July 17, the company rolled out a batch of video updates for its Firefly generative hub, headlined by a feature that turns your vocal impressions into synced sound effects.

Beatboxing your Foley effects

Here is how it works: you type a quick prompt describing the sound you need, and then you record yourself making the noise into your microphone.

In a live demo, an Adobe employee added a zipper sound by simply making a sharp "zzzztttt" noise into the mic. Firefly took that cue and generated a matching zipper effect at the exact volume and timing. It sounded convincing, though footsteps on concrete proved a bit trickier. Adobe clarified that the model is not mimicking your vocal tone; it just uses your recording to gauge intensity and pacing. A timeline editor sits right at the bottom so you can lock the sound to the cut.

Alongside audio, Adobe launched three new visual controls. Composition Reference lets you guide generation with an existing photo or video. Video Presets let you lock into anime, black-and-white, or vector styles. Keyframe Cropping lets you upload the opening and closing frames with a fixed aspect ratio so the video stays tightly in frame.

Bringing in Google's Veo 3 and tighter controls

Adobe has been dropping major Firefly updates roughly every month since redesigning the app in April. After opening the door to third-party models in June, the company just added Google's Veo 3, first shown off at Google I/O 2025 back in May.

Veo 3 is one of the few frontier models capable of producing video with synchronized sound natively. To keep enterprise clients calm, Google agreed not to train its future models on Adobe user data. Every clip created in Firefly also gets digitally signed with the specific model used, giving commercial artists an audit trail against copyright risk. Zeke Koch, Firefly's VP of product management, noted the team is relentlessly shipping and will onboard more models as long as partners sign off on Adobe's privacy terms.

Why it matters

For solo creators and animators, this is worth testing immediately. Foley work is tedious, and being able to vocalize the exact beat of an impact or swoosh solves the frustrating timing mismatch common in AI video ideation.

Across the industry, the AI video race is clearly moving past standalone visual prompts into granular directors' controls and audio-visual synchronization. Just delivering a five-second silent MP4 does not cut it anymore for commercial pipelines.

We think Adobe's playbook here is very clever. Instead of burning billions trying to beat Google at training raw foundation video models, Adobe simply brought Veo 3 inside its walls. By wrapping external models with copyright guardrails, timeline editors, and intuitive audio cues, Adobe positions itself as the enterprise safe harbor. The only risk is reliance: if third-party models handle the heaviest lifting, Adobe must continually prove that its interface and workflow wrapper provide enough long-term value to keep creatives paying.