At its annual MAX conference, Adobe took another step toward building a true all-in-one creative AI studio. The company announced major upgrades to Adobe Firefly, expanding it beyond image generation into a full multimedia creation space that now includes AI-powered audio, video, and voice tools — all backed by a growing lineup of partner models and Adobe’s own generative tech.

Among the biggest updates is Generate Soundtrack, which lets creators instantly produce royalty-free music synced to their visuals, and Generate Speech, a text-to-voice tool with natural delivery and multilingual support. Firefly is also introducing a timeline-based AI video editor, combining traditional multitrack editing with prompt-driven generation, so users can mix recorded clips with new, AI-made visuals or scenes in the same workspace.

Adobe also previewed Project Moonlight, a conversational AI interface designed to work across its creative apps. It can analyze a user’s projects and social posts, then suggest new content ideas or automate repetitive creative tasks — a move that signals Adobe’s deeper investment in “agentic AI,” where assistants don’t just generate assets but help shape creative direction.

The new Firefly Image Model 5 continues Adobe’s focus on high-fidelity visual generation, now capable of 4MP native resolution and improved photorealism. The company has also opened Firefly to third-party models from ElevenLabs, Google, Runway, Luma AI, and Topaz Labs, among others, letting users pick the right model for each task.

For creators who want more control, Firefly Custom Models are now in private beta, allowing individuals or teams to train their own personalized AI models that replicate their unique aesthetic or brand style — all within Adobe’s ecosystem.

Firefly’s new tools aim to streamline the creative process from brainstorming to publishing, letting users ideate, edit, and finish projects in one space. The updates are rolling out in phases, with most features — including Generate Soundtrack and Firefly Image Model 5 — already in public beta.